Search method, search program, molded product, and manufacturing method

Abstract

To provide a search method, search program, molded product, and manufacturing method that can obtain appropriate conditions in a short time. [Solution] The search method according to the present invention includes: a three-dimensional model creation step of creating a three-dimensional model based on design parameters; a simulation step of performing a fluid simulation using the three-dimensional model; a shape parameter extraction step of extracting shape parameters by image analysis; a convergence determination step of determining whether the difference based on the extracted shape parameters is below a convergence threshold; an update step of updating the design parameters by an optimization algorithm; a repeating step of performing the three-dimensional model creation step, simulation step, shape parameter extraction step and convergence determination step for the design parameters updated by the update step; and an output step of outputting the design parameters to be determined as optimal design parameters.

Description

Technical Field

[0001] The present invention relates to a search method, a search program, a molded article, and a manufacturing method.

Background Art

[0002] In recent years, in material design, as a method for searching for design conditions to obtain desired properties, a technique for searching design conditions using Bayesian optimization has been disclosed (see, for example, Patent Documents 1 and 2). For example, when designing the cross-sectional shape of a fiber, conditions such as the hole shape of a die through which a liquid resin passes, temperature, and resin flow rate are designed.

Prior Art Documents

Non-Patent Documents

[0003]

Non-Patent Document 1

Non-Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0004] By the way, in recent years, with the diversification of needs and the sophistication of design technology, the range and items of conditions that can be designed, such as the cross-sectional shape of fibers and raw material types, have increased, and the time required for searching has been increasing.

[0005] The present invention has been made in view of the above, and an object thereof is to provide a search method, a search program, a molded article, and a manufacturing method capable of obtaining appropriate conditions in a short time.

Means for Solving the Problems

[0006] To solve the above-mentioned problems and achieve the objective, the search method according to the present invention is a search method in which a computer searches for optimal design parameters for obtaining a molded product having a predetermined material shape, and includes: an initial setting step of setting design parameters for manufacturing a molded product and a convergence threshold; a three-dimensional model creation step of creating a three-dimensional model based on the design parameters; a simulation step of executing a fluid simulation using the three-dimensional model; a shape parameter extraction step of extracting shape parameters by image analysis of the simulation results; a convergence determination step of determining whether the difference between the shape parameters extracted in the shape parameter extraction step and the shape parameter / target value extracted in the extraction step executed immediately before is less than or equal to the convergence threshold; an update step of updating the design parameters by an optimization algorithm using a set of all extracted shape parameters and design parameters if it is determined that the difference is greater than the convergence threshold; a repeating step of executing the three-dimensional model creation step, the simulation step, the shape parameter extraction step and the convergence determination step for the design parameters updated by the update step; and an output step of outputting the design parameters to be determined as optimal design parameters if it is determined that the difference is less than or equal to the convergence threshold.

[0007] Furthermore, the search method according to the present invention further includes a simulation count determination step in which it is determined whether or not the simulation in the search process for the molded product is the second or subsequent simulation, and the convergence determination step is performed when it is determined that the simulation is the second or subsequent simulation.

[0008] Furthermore, in the search method according to the present invention, the update step updates the design parameters by Bayesian optimization.

[0009] Furthermore, the search method according to the present invention includes, in the above invention, an update step of determining whether the number of updates of the best value of the shape parameter is equal to or greater than a preset number of updates; a modification step of changing the hyperparameters or acquisition function by Bayesian optimization if it is determined that the number of consecutive updates of the best value of the shape parameter is less than the preset number; a design parameter update step of updating the design parameter by Bayesian optimization using the hyperparameters or acquisition function used in the previous update if it is determined that the number of consecutive updates of the best value of the shape parameter is equal to or greater than the preset number, and updating the design parameter by Bayesian optimization using the hyperparameters or acquisition function changed in the modification step if it is determined that the number of consecutive updates of the best value of the shape parameter is less than the preset number.

[0010] Furthermore, the search method according to the present invention includes, in the above invention, an update step of determining whether the number of consecutive updates of the best value of the shape parameter is equal to or greater than a preset number of updates; a modification step of changing the hyperparameters of the optimization algorithm if it is determined that the number of consecutive updates of the best value of the shape parameter is less than the preset number of updates; a design parameter update step of updating the design parameter with the optimization algorithm using the hyperparameters used in the previous update if it is determined that the number of consecutive updates of the best value of the shape parameter is equal to or greater than the preset number of updates, and updating the design parameter with the optimization algorithm using the hyperparameters changed in the modification step if it is determined that the number of consecutive updates of the best value of the shape parameter is less than the preset number of updates.

[0011] Furthermore, the search method according to the present invention includes, in the above invention, a simulation step of a low-fidelity fluid simulation step of performing a low-fidelity fluid simulation using the three-dimensional model; a provisional extraction step of extracting provisional shape parameters by image analysis of the low-fidelity fluid simulation results; a search and selection step of searching and selecting candidates from the extracted provisional shape parameters to be used in a high-fidelity fluid simulation which has a higher computational load than the low-fidelity fluid simulation; and a high-fidelity fluid simulation step of performing the high-fidelity fluid simulation using the three-dimensional model based on the design parameters selected in the search and selection step.

[0012] Furthermore, in the search method according to the present invention, the search and selection step involves selecting a predetermined number of design parameters, in which the values ​​of the acquisition function based on the low-fidelity fluid simulation results are among the highest, as candidates to be used in the high-fidelity simulation.

[0013] Furthermore, the search method according to the present invention further includes a candidate / selection number update step, which, before executing the iterative step, reduces the number of candidates to be selected in the low-fidelity fluid simulation and the high-fidelity fluid simulation from an initial value.

[0014] Furthermore, the search program according to the present invention is a search program that causes a computer to search for optimal design parameters for obtaining a molded product having a predetermined material shape, and causes the computer to execute the following: an initial setting step of setting design parameters for manufacturing a molded product and a convergence threshold; a three-dimensional model creation step of creating a three-dimensional model based on the design parameters; a simulation step of executing a fluid simulation using the three-dimensional model; a shape parameter extraction step of extracting shape parameters by image analysis of the simulation results; a convergence determination step of determining whether the difference between the shape parameters extracted in the shape parameter extraction step and the shape parameters / target value extracted in the extraction step executed immediately before is less than or equal to the convergence threshold; an update step of updating the design parameters by an optimization algorithm using a set of all extracted shape parameters and design parameters if it is determined that the difference is greater than the convergence threshold; a repeating step of executing the three-dimensional model creation step, the simulation step, the shape parameter extraction step and the convergence determination step for the design parameters updated in the update step; and an output step of outputting the design parameters to be determined as optimal design parameters if it is determined that the difference is less than or equal to the convergence threshold.

[0015] Furthermore, the molded product according to the present invention also relates to a molded product manufactured using design parameters obtained by the search method according to the above invention. Specifically, the present invention provides a molded product manufactured using design parameters obtained by the search method described in any one of claims 1 to 5.

[0016] Furthermore, the manufacturing method according to the present invention manufactures a molded product using design parameters obtained by the exploration method according to the above invention. [Effects of the Invention]

[0017] According to the present invention, appropriate conditions can be obtained in a short time when searching for design conditions for fibers.

Brief Description of the Drawings

[0018] [Figure 1] FIG. 1 is a diagram showing a schematic configuration of a search system according to an embodiment of the present invention. [Figure 2] FIG. 2 is a block diagram showing a configuration of a search device included in the search system according to an embodiment of the present invention. [Figure 3] FIG. 3 is a flowchart showing an outline of a design search process performed by the search device according to an embodiment of the present invention. [Figure 4] FIG. 4 is a partial cross-sectional view for explaining a die used for generating fibers. [Figure 5] FIG. 5 is a plan view showing the die as viewed from the direction of arrow A shown in FIG. 4. [Figure 6] FIG. 6 is a diagram showing an example of a three-dimensional model of a die used for generating fibers. [Figure 7] FIG. 7 is a flowchart showing an outline of a design search process performed by the search device according to Modification 1 of the present invention. [Figure 8] FIG. 8 is a flowchart showing an outline of a design search process performed by the search device according to Modification 2 of the present invention. [Figure 9] FIG. 9 is a flowchart showing an outline of a design search process performed by the search device according to Modification 3 of the present invention. [Figure 10] FIG. 10 is a flowchart showing an outline of a design search process performed by the search device according to Modification 4 of the present invention. [Figure 11] FIG. 11 is a flowchart showing an outline of a design search process performed by the search device according to Modification 5 of the present invention.

Embodiments for Carrying Out the Invention

[0019] The following describes in detail, with reference to the drawings, embodiments of the exploration method, exploration program, molded product, and manufacturing method according to the present invention. However, the present invention is not limited to these embodiments. Furthermore, the individual embodiments of the present invention are not independent but can be combined and implemented as appropriate.

[0020] (Embodiment) [System Configuration] Figure 1 is a diagram showing the schematic configuration of a search system according to an embodiment of the present invention. The search system 1 shown in these figures comprises a search device 2 for designing the cross-sectional shape of a fiber exhibiting set characteristics, a display device 3 for displaying various information including the design information of the search device 2, and an input device 4.

[0021] The design information generated by the search device 2 includes, for example, the raw materials that make up the fiber, the cross-sectional shape, the properties of the material, the type of die, and manufacturing conditions such as temperature and flow rate. Based on the initial settings parameters set in advance, the search device 2 searches for setting parameters that result in stable shape parameters for the estimated fiber shape and designs the optimal design parameters. The following description explains an example in which the design parameters of a die used in fiber production are used to obtain the optimal design parameters from the shape parameters of the fiber produced by the die.

[0022] Figure 2 is a block diagram showing the configuration of the search device included in the search system according to an embodiment of the present invention. The search device 2 includes a three-dimensional model creation unit 21, an analysis unit 22, a shape parameter extraction unit 23, a determination unit 24, an optimization unit 25, a control unit 26, and a storage unit 27.

[0023] The 3D model creation unit 21 generates a 3D model of the jaw using design parameters. The 3D model creation unit 21 can use known methods, for example, SolidWorks, Fusion360, FreeCAD, etc. to generate the 3D model.

[0024] The analysis unit 22 analyzes the flow of material passing through the nozzle based on material information, manufacturing conditions, and the nozzle shape generated by the three-dimensional model creation unit 21. Material information includes material type and viscosity. Manufacturing conditions include environmental conditions during manufacturing, such as temperature and pressure. The analysis unit 22 analyzes the material flow using computational fluid dynamics (CFD). The analysis unit 22 can use known methods, such as Ansys Fluent, OpenFOAM, Simcenter STAR-CCM+, etc.

[0025] The shape parameter extraction unit 23 extracts shape parameters indicating the shape of the fiber using the analysis results from the analysis unit 22. The shape parameter extraction unit 23 analyzes the image (fiber shape image) obtained from the analysis unit 22 to extract the shape parameters. The shape parameters are parameters that indicate the characteristic quantities of the fiber, including, for example, the area (e.g., cross-sectional area) of a cross-section with a plane perpendicular to the longitudinal direction of the fiber as the cutting plane, the outer circumference, the fiber surface area, the circumscribed rectangle, the aspect ratio, the circularity, and any other parameters that represent the shape. It is preferable that the shape parameters are selected according to the set characteristics. The shape parameter extraction unit 23 can use known image analysis methods, such as MATLAB®, ImageJ, OpenCV, etc.

[0026] The determination unit 24 determines whether the number of simulations performed by the analysis unit 22 and the difference (residual) based on the shape parameters satisfy the pre-set conditions.

[0027] The optimization unit 25 updates the design parameters by updating them using an optimization algorithm. Examples of optimization algorithms include Bayesian optimization, genetic algorithms, and simulated annealing. In this embodiment, an example of updating the design parameters using Bayesian optimization is described, but other optimization algorithms may also be used.

[0028] The control unit 26 comprehensively controls the operation of the search device 2.

[0029] The three-dimensional model creation unit 21, analysis unit 22, shape parameter extraction unit 23, determination unit 24, optimization unit 25, and control unit 26 are computers configured using one or more hardware components such as a CPU (Central Processing Unit), GPU (Graphics Processing Unit), ASIC (Application Specific Integrated Circuit), and FPGA (Field Programmable Gate Array). Furthermore, the search device 2 is connected to the display device 3 and input device 4 via a communication network, enabling it to send and receive information. The communication network referred to here is configured using, for example, an existing public telephone network, LAN (Local Area Network), WAN (Wide Area Network), etc., and can be wired or wireless.

[0030] The memory unit 27 stores data including various programs for operating the search device 2, and various parameters necessary for the operation of the search device 2. The various programs include a search program for searching for material design conditions to obtain a desired material shape (in this case, a fiber shape).

[0031] The memory unit 27 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.

[0032] 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.

[0033] The display device 3 is a display made of liquid crystal or organic EL (Electro-Luminescence), and is electrically connected to the search device 2. Under the control of the display control unit, the display device 3 acquires and displays display data output from the search device 2. The display device 3 may also have an audio output function such as a speaker.

[0034] Input device 4 accepts various types of information, including settings related to the search process, and outputs the received information to search device 2. Input device 4 is configured using a user interface such as a keyboard, mouse, microphone, or touch panel.

[0035] Next, the design search process performed by the search device will be explained with reference to Figures 3 to 6. Figure 3 is a flowchart outlining the design search process performed by a search device according to one embodiment of the present invention. When the control unit 26 receives an instruction to execute the design search process, it sets the initial design parameters x and the convergence threshold (step S101: initial setting step). The control unit 26 sets the initial design parameters by, for example, reading the information input by the input device 4 or the initial design parameters and convergence threshold from the storage unit 27. The convergence threshold is a threshold for determining whether or not to perform optimization, and is set for the difference (residual) between the shape parameters before and after optimization.

[0036] Here, the die, which is expressed as a design parameter, will be explained with reference to Figures 4 and 5. Figure 4 is a partial cross-sectional view illustrating the die used to generate fibers. Figure 5 is a plan view showing the die as seen from the direction A indicated by the arrow in Figure 4. The die 100 shown in Figure 4 has a tapered conical shape with a plurality of holes 101a formed that penetrate from one end to the other in the axial direction. The die 100 has a material inlet surface 101 on the conical bottom side, and a material discharge section 102 is provided on the opposite side of the inlet surface 101 (the conical tip side). As shown in Figure 5, openings 101b of the holes 101a are formed in the inlet surface 101. The shape of these openings 101b, the arrangement of each opening 101b, the path of the holes 101a and their length determine the shape of the material discharged from the discharge section 102 (corresponding to the cross-section of the fiber). Design parameters are a set of parameters that express the shape, number, and arrangement of holes in such nozzles.

[0037] Returning to Figure 3, after setting the initial design parameters and convergence threshold, the 3D model creation unit 21 creates a 3D model based on the design parameters (Step S102: 3D model creation step).

[0038] Figure 6 shows an example of a three-dimensional model of a nozzle used to generate fibers. Figure 6 shows an example of a three-dimensional model created with FreeCAD. The three-dimensional model creation unit 21 creates a three-dimensional model of the nozzle based on the design parameters using the method described above. For example, a nozzle model 200 as shown in Figure 6 is created based on the design parameters. Figure 6 shows a perspective view from the inlet surface 201 side and has a mesh structure corresponding to the shape. The model 200 also has a hole 201a through which the material flows, and in Figure 6, the opening 201b of the hole 201a through which the material flows is shown as a black fill, depending on the design parameters.

[0039] Returning to Figure 3, the analysis unit 22 performs a CFD simulation using the created three-dimensional model (step S103: simulation step). As described above, the analysis unit 22 analyzes the flow of material through the nozzle. For example, as shown in Figure 6, the material flows in the direction D F The material flow when flowing through the nozzle (hole 201a in model 200) will be analyzed using OpenFOAM.

[0040] After the analysis process, the shape parameter extraction unit 23 extracts the shape parameter y of the fibers formed via the die based on the analysis results (step S104: shape parameter extraction step). As described above, the analysis unit 22 extracts the shape features of the fibers from the analysis results of the material flow through the die, and uses these as the shape parameter y.

[0041] Then, the determination unit 24 determines whether the simulation performed in step S103 is the second or subsequent simulation (step S105). The determination unit 24 reads the number of simulations performed immediately before in the design search process and determines whether it is the second or subsequent simulation. If the determination unit 24 determines that the simulation is not the second or subsequent simulation (it is the first simulation) (step S105: No), the control unit 26 proceeds to step S107. Conversely, if the determination unit 24 determines that the simulation is the second or subsequent simulation (step S105: Yes), the control unit 26 proceeds to step S106.

[0042] In step S106, the determination unit 24 calculates the residual of the shape parameter y and determines whether the residual is less than or equal to the convergence threshold (convergence determination step). The determination unit 24 calculates the difference between the shape parameter y calculated in step S104 and the shape parameter y calculated in the immediately preceding process and determines whether it is less than or equal to the convergence threshold set in step S101. If the determination unit 24 determines that the residual is less than or equal to the convergence threshold (step S106: Yes), the control unit 26 proceeds to step S108. Conversely, if the determination unit 24 determines that the residual is not less than or equal to the convergence threshold (step S106: No), the control unit 26 proceeds to step S107. In this case, the residual may be the difference between specified parameters, or, if calculated using multiple parameters, the mean, mode, median, maximum, or minimum of the residuals of the same type of parameter may be used. Furthermore, when calculating residuals using multiple types of parameters, examples include the sum or product of the residuals of the same type of parameter, or a weighted average according to the type of parameter (for example, if there are two types of parameters, A and B, and parameter A is given twice as much weight as parameter B, then (2A+B) / 3 would be used). It is possible to set which value to treat as the residual.

[0043] In step S107, the optimization unit 25 updates the design parameter x by Bayesian optimization (update step). In step S107, a new design parameter x is generated by Bayesian optimization based on the design parameter and all the shape parameters y extracted in step S104 for that design parameter. After the design parameters are updated by the optimization unit 25, the control unit 26 returns to step S102 and repeats the process described above for the updated design parameters (repeat step). In this embodiment, the design parameter update process performed by the optimization unit 25 corresponds to the update step.

[0044] Furthermore, when the residual falls below the convergence threshold, the control unit 26 outputs the latest design parameter x as the optimal design parameter (step S108: output step). The output destinations at this time are the storage unit 27 and the display device 3. When the optimal design parameter is input to the storage unit 27, the storage unit 27 stores the input optimal design parameter. Also, when the optimal design parameter is input to the display device 3, the display device 3 displays the input optimal design parameter.

[0045] As described above, once the optimal design parameters are set, a molded product is manufactured based on those design parameters. In this process, a die is manufactured as the molded product corresponding to the design parameters. Then, under the set raw materials and manufacturing conditions such as temperature and flow rate, fibers are manufactured using this die.

[0046] In the embodiment described above, a three-dimensional model of the nozzle is created from the design parameters, shape parameters representing the predicted fiber shape (cross-sectional shape) are extracted from the results of analyzing the material flow using the three-dimensional model, and optimal design parameters are set based on the residuals with the shape parameters extracted immediately before. According to this embodiment, by combining CFD simulation to analyze fluid flow, parameterization of analysis results by image analysis, and generation of design parameters by Bayesian optimization, appropriate conditions can be obtained in a short time when searching for design conditions for fibers.

[0047] The design method according to this embodiment is particularly useful in designs where there is a large gap between the type of fiber and its function, such as in clothing, and where there are many parameters, including fiber type, fiber cross-section, and the shape of the die for producing the fiber.

[0048] (Variation 1) Next, a modified example 1 of this embodiment will be described. The search system according to this modified example 1 has the same configuration as the search system 1 according to the embodiment, so its description will be omitted. The following describes the differences from the embodiment. In this modified example 1, the optimization process of the design parameters is different from that of the embodiment described above.

[0049] Figure 7 is a flowchart showing an overview of the design search process performed by the search device according to Modification 1 of the present invention. In this Modification 1, when the search device 2 receives an instruction to execute the design search process, it performs the following actions in the same manner as steps S101 to S105 in Figure 3: creation of a three-dimensional model based on design parameters x, execution of CFD simulation, extraction of shape parameters y, and determination of the number of simulations (steps S201 to S205). In step S205, if the determination unit 24 determines that the simulation is not the second or later (step S205: No), the control unit 26 proceeds to step S209. Conversely, if the determination unit 24 determines that the simulation is the second or later (step S205: Yes), the control unit 26 proceeds to step S206.

[0050] In step S206, the determination unit 24 calculates the residual of the shape parameter y in the same manner as in step S106 and determines whether the residual is less than or equal to the convergence threshold. The determination unit 24 calculates the difference between the shape parameter y calculated in step S204 and the shape parameter y calculated in the immediately preceding process and determines whether it is less than or equal to the convergence threshold set in step S201. If the determination unit 24 determines that the residual is less than or equal to the convergence threshold (step S206: Yes), the control unit 26 proceeds to step S210. Conversely, if the determination unit 24 determines that the residual is not less than or equal to the convergence threshold (step S206: No), the control unit 26 proceeds to step S207.

[0051] In step S207, the determination unit 24 determines whether the best value of the shape parameter y has been updated a predetermined number of times consecutively. Here, the "best value" in this modified example 1 is a value from the group of shape parameters y obtained from the simulation that satisfies predetermined conditions. The predetermined conditions are set according to the type of shape parameter, and include, for example, the maximum value, minimum value, average value, mode value, etc. This best value is updated if the shape parameter y obtained by a subsequent new CFD simulation satisfies the conditions favorably. If the determination unit 24 determines that the best value has been updated a predetermined number of times consecutively (step S207: Yes), the control unit 26 proceeds to step S209. Conversely, if the determination unit 24 determines that the best value has not been updated a predetermined number of times consecutively (step S207: No), the control unit 26 proceeds to step S208.

[0052] In step S208, the optimization unit 25 changes the hyperparameters or acquisition function in Bayesian optimization. For example, the optimization unit 25 may change the acquisition function currently used, such as PI (Probability of Improvement), PTR (Probability in Target Range), EI (Expected Improvement), or UCB / LCB (Upper / Lower Confidence Bound), to another acquisition function, or it may change the hyperparameters of the Bayesian optimization algorithm used in UCB / LCB.

[0053] In step S209, the optimization unit 25 updates the design parameters x by Bayesian optimization in the same manner as in step S107. If the hyperparameters or acquisition function were changed in step S208, the design parameters x are updated using the changed hyperparameters or acquisition function. After the design parameters are updated by the optimization unit 25, the control unit 26 returns to step S202 and repeats the above-described process for the updated design parameters. In this modified example 1, steps S207 to S209, which are the design parameter update process, correspond to the update step.

[0054] Furthermore, if the residual falls below the convergence threshold, the control unit 26 outputs the latest design parameter x as the optimal design parameter (step S210).

[0055] In Modification 1 described above, similar to the embodiment, a three-dimensional model of the nozzle is created from the design parameters, and shape parameters representing the predicted fiber shape (cross-sectional shape) are extracted from the results of analyzing the material flow using the three-dimensional model. The optimal design parameters are then set based on the residuals with the shape parameters extracted immediately before. According to this Modification 1, by combining CFD simulation to analyze fluid flow, parameterization of analysis results by image analysis, and generation of design parameters by Bayesian optimization, appropriate conditions can be obtained in a short time when searching for design conditions for fibers.

[0056] Furthermore, according to this modified example 1, the hyperparameters or acquisition function in Bayesian optimization are changed according to the continuous update pattern of the best value of the shape parameter y. This makes it easier to escape even if a local minimum is encountered, and allows for the search for optimal design parameters.

[0057] (Modification 2) Next, a modified example 2 of this embodiment will be described. The search system in this modified example 2 has the same configuration as the search system 1 in the embodiment, so its description will be omitted. The following describes the differences from the embodiment. In this modified example 2, the optimization process of the design parameters is different from that of the embodiment described above.

[0058] Figure 8 is a flowchart showing an overview of the design search process performed by the search device according to Modification 2 of the present invention. In this Modification 2, when the search device 2 receives an instruction to execute the design search process, it performs the following actions in the same manner as steps S101 to S105 in Figure 3: creation of a three-dimensional model based on design parameters x, execution of a CFD simulation, extraction of shape parameters y, and determination of the number of simulations (steps S301 to S305). If the determination unit 24 determines that the simulation is not the second or later (step S305: No), the control unit 26 proceeds to step S310. Conversely, if the determination unit 24 determines that the simulation is the second or later (step S305: Yes), the control unit 26 proceeds to step S306.

[0059] In step S306, the determination unit 24 determines whether the residual is less than or equal to the convergence threshold. If the determination unit 24 determines that the residual is less than or equal to the convergence threshold (step S306: Yes), the control unit 26 proceeds to step S311. Conversely, if the determination unit 24 determines that the difference is not less than or equal to the convergence threshold (step S306: No), the control unit 26 proceeds to step S307.

[0060] In step S307, the determination unit 24 determines whether the best value of the shape parameter y has been updated a predetermined number of times consecutively. If the determination unit 24 determines that the best value has been updated a predetermined number of times consecutively (step S307: Yes), the control unit 26 proceeds to step S310. Conversely, if the determination unit 24 determines that the best value has not been updated a predetermined number of times consecutively (step S307: No), the control unit 26 proceeds to step S308.

[0061] In step S308, the optimization unit 25 modifies the hyperparameters related to the Gaussian process regression model. For example, the optimization unit 25 modifies the hyperparameters of the Gaussian process regression model. This modification changes the optimization method. The hyperparameters here are kernel functions. Examples of kernel functions include the Gaussian (Radial Basis Function: RBF) kernel, Matern kernel, Constant kernel, and White kernel, and any combination of these may be used. Note that if the optimization is for initial setting parameters (initial design parameters x), this step S308 may be omitted.

[0062] In step S309, the optimization unit 25 constructs an optimization algorithm using the modified hyperparameters.

[0063] In step S310, the optimization unit 25 updates the design parameters x using the optimization algorithm. If the hyperparameters were changed in step S308, the optimization algorithm is updated using the changed hyperparameters to update the design parameters x. After the design parameters are updated by the optimization unit 25, the control unit 26 returns to step S302 and repeats the above-described process for the updated design parameters. In this modified example 2, steps S307 to S310, which are the design parameter update processes, correspond to the update steps.

[0064] Furthermore, if the residual falls below the convergence threshold, the control unit 26 outputs the latest design parameter x as the optimal design parameter (step S311).

[0065] In the modified example 2 described above, similar to the embodiment, a three-dimensional model of the nozzle is created from the design parameters, and shape parameters representing the predicted fiber shape (cross-sectional shape) are extracted from the results of analyzing the material flow using the three-dimensional model. The optimal design parameters are then set based on the residuals with the shape parameters extracted immediately before. According to this modified example 2, by combining CFD simulation to analyze fluid flow, parameterization of analysis results by image analysis, and generation of design parameters by Bayesian optimization, appropriate conditions can be obtained in a short time when searching for design conditions for fibers.

[0066] Furthermore, according to this modified example 2, the hyperparameters in the optimization algorithm are changed according to the continuous update pattern of the best value of the shape parameter y. Therefore, design parameters can be searched using an optimization algorithm other than Bayesian optimization, and the optimal design parameters can be searched without getting stuck in a local minimum.

[0067] (Variation 3) Next, a third modification of this embodiment will be described. The search system in this third modification has the same configuration as the search system 1 in the embodiment, so its description will be omitted. The following describes the differences from the embodiment. In this third modification, the content of the determination process for whether or not to perform design parameter optimization processing differs from that of the embodiment described above.

[0068] Figure 9 is a flowchart outlining the design search process performed by the search device according to Modification 3 of the present invention. In this Modification 3, when the search device 2 receives an instruction to execute the design search process, it performs the following actions in the same manner as steps S101 to S104 in Figure 3: creation of a three-dimensional model based on design parameters x, execution of CFD simulation, and extraction of shape parameters y (steps S401 to S404). In this Modification 3, in step S401, in addition to the initial design parameters and convergence threshold, a target value z for the shape parameter y is set. At this time, the convergence threshold is set to the value of the difference between the shape parameter y and the target value z. In step S405, the determination unit 24 calculates the difference between the shape parameter y and the target value z, and determines whether the difference is less than or equal to the convergence threshold. If the determination unit 24 determines that the difference is less than or equal to the convergence threshold (step S405: Yes), the control unit 26 proceeds to step S407. Conversely, if the determination unit 24 determines that the difference is not less than or equal to the convergence threshold (step S405: No), the control unit 26 proceeds to step S406.

[0069] In step S406, the optimization unit 25 updates the design parameter x by Bayesian optimization in the same manner as in step S107 (update step). After the design parameters are updated by the optimization unit 25, the control unit 26 returns to step S402 and repeats the above-described process for the updated design parameters (repeat step).

[0070] Furthermore, if the difference falls below the convergence threshold, the control unit 26 outputs the latest design parameter x as the optimal design parameter (step S407).

[0071] In the modified example 3 described above, a three-dimensional model of the nozzle is created from the design parameters, and shape parameters representing the predicted fiber shape (cross-sectional shape) are extracted from the results of analyzing the material flow using the three-dimensional model. The optimal design parameters are then set based on the difference from the target value. According to this modified example 3, similar to the embodiment, by combining CFD simulation to analyze the fluid flow, parameterization of the analysis results by image analysis, and generation of design parameters by Bayesian optimization, appropriate conditions can be obtained in a short time when searching for design conditions for the fibers.

[0072] Furthermore, in this modified example 3, the design parameter update process related to modified examples 1 and 2 can be applied.

[0073] (Modification 4) Next, a modified example 4 of this embodiment will be described. The search system in this modified example 4 has the same configuration as the search system 1 in the embodiment, so its description will be omitted. The following describes the differences from the embodiment. In this modified example 4, the CFD simulation process differs from the embodiment described above.

[0074] Figure 10 is a flowchart outlining the design search process performed by the search device according to Modification 4 of the present invention. In this Modification 4, when the search device 2 receives an instruction to execute the design search process, it creates a three-dimensional model based on the design parameters x in the same manner as in steps S101 and S102 of Figure 3 (steps S501 and S502). At this time, in step S501 (initial setting step), the control unit 26 sets the number of design parameters x for the nozzle to be executed in the low-fidelity CFD simulation and high-fidelity CFD simulation described later.

[0075] After creating the three-dimensional model, the analysis unit 22 performs a CFD simulation using the created three-dimensional model (steps S503 to S506: simulation steps). The analysis unit 22 analyzes the flow of material through the nozzle. For example, as shown in Figure 6, the material flows in the flow direction D F The material flow when flowing through the nozzle (hole 201a in model 200) will be analyzed using OpenFOAM.

[0076] In the simulation step, the analysis unit 22 performs a low-fidelity CFD simulation and a high-fidelity CFD simulation, which has a higher computational load than the low-fidelity CFD simulation. Here, the low-fidelity CFD simulation and the high-fidelity CFD simulation have the following differences in execution parameters. • Mesh (grid-related parameters) Mesh resolution: The number of cells (elements) is approximately 10 to 100 times greater (low-fidelity CFD < high-fidelity CFD). Mesh density gradient (local mesh refinement): Low-fidelity CFD < High-fidelity CFD Mesh type (polyhedral mesh): Low-fidelity CFD < High-fidelity CFD • Time discretization parameters Time step size (Δt): Low-fidelity CFD > High-fidelity CFD A smaller Δt indicates higher accuracy and longer computation time. • Time integration scheme First-order precision vs. second-order precision: High-fidelity CFD uses higher-order schemes. Implicit vs. Explicit Methods: Implicit methods are used in high-fidelity CFD. Implicit methods are more stable but take longer to compute. Number of iterations (number of internal iterations at each time step): Low-fidelity CFD < High-fidelity CFD • Convergence criteria Threshold for residuals of physical quantities used in the determination: Low-fidelity CFD > High-fidelity CFD

[0077] The analysis unit 22 first performs a low-fidelity CFD simulation (step S503: low-fidelity fluid simulation step). This low-fidelity CFD simulation analyzes the flow of material passing through the nozzle. In this process, the analysis unit 22 performs low-fidelity CFD simulations for multiple nozzle design parameters.

[0078] After the analysis process, the shape parameter extraction unit 23 extracts a provisional shape parameter y' of the fiber formed via the die based on the analysis results (step S504: provisional extraction step). As described above, the analysis unit 22 extracts the shape features of the fiber from the analysis results of the material flow through the die, and uses this as the provisional shape parameter y'.

[0079] The analysis unit 22 searches for and selects candidates to be used in high-fidelity CFD simulation from the extracted provisional shape parameters y' (step S505: search and selection step). The analysis unit 22 searches for and selects a predetermined number of design parameters x that are ranked highly, using, for example, an evaluation score obtained by inputting the provisional shape parameters y' into a trained model constructed by Bayesian optimization, an acquisition function obtained by Bayesian optimization, and a predetermined target value.

[0080] Then, the analysis unit 22 performs a high-fidelity CFD simulation using the selected candidates (step S506: high-fidelity fluid simulation step). This high-fidelity CFD simulation analyzes the flow of material passing through the nozzle in a three-dimensional model based on the selected candidates (design parameter x).

[0081] After the analysis process, the shape parameter extraction unit 23 extracts the shape parameter y of the fibers formed via the die based on the analysis results from the high-fidelity CFD simulation (step S507: shape parameter extraction step). As described above, the analysis unit 22 extracts the shape features of the fibers from the analysis results of the material flow through the die, and uses these as the shape parameter y.

[0082] After extracting the shape parameter y, the search device 2 outputs the optimal design parameters in the same manner as steps S105 to S108 described above (steps S508 to S511).

[0083] In the modified example 4 described above, similar to the embodiment, a three-dimensional model of the nozzle is created from the design parameters, and shape parameters representing the predicted fiber shape (cross-sectional shape) are extracted from the results of analyzing the material flow using the three-dimensional model. The optimal design parameters are then set based on the residuals with the shape parameters extracted immediately before. According to this modified example 4, by combining CFD simulation to analyze fluid flow, parameterization of analysis results by image analysis, and generation of design parameters by Bayesian optimization, appropriate conditions can be obtained in a short time when searching for design conditions for fibers.

[0084] Furthermore, according to this modified example 4, by performing low-fidelity CFD simulations and high-fidelity CFD simulations, it is possible to efficiently search for the optimal solution while suppressing the computational load by combining multiple evaluation models with different accuracies (fidelity). In this modified example 4, by performing low-fidelity CFD simulations and high-fidelity CFD simulations, the computation time can be reduced by approximately 10 to 100 times compared to the processing according to the embodiment.

[0085] (Variation 5) Next, a modified example 5 of this embodiment will be described. The search system in this modified example 5 has the same configuration as the search system 1 in the embodiment, so its description will be omitted. The following describes the differences from the embodiment and modified example 4. In this modified example 5, the processing in the iterative step is different from the design search process in modified example 4 described above.

[0086] Figure 11 is a flowchart showing an overview of the design search process performed by the search device according to Modification 5 of the present invention. In this Modification 5, when the search device 2 receives an instruction to execute the design search process, it performs the three-dimensional model creation process, CFD simulation process, shape parameter y extraction process, and determination process (steps S601 to S609) in the same manner as steps S501 to S509 in Figure 10.

[0087] Furthermore, in step S610, the optimization unit 25 updates the design parameter x by Bayesian optimization (update step).

[0088] In this modified example 5, the optimization unit 25 updates the number of candidates and the number of selections when performing the CFD simulation during the iterative steps (step S611: Candidate / Selection Count Update Step). In step S611, the optimization unit 25 reduces the number of candidates and the number of selections relative to the initial settings, for example, according to the number of iterative steps performed.

[0089] After the design parameters and the number of candidates / selections are updated by the optimization unit 25, the control unit 26 returns to step S602 and repeats the above-described process for the updated design parameters (repeated step). In this modified example 5, the CFD simulation process in steps S603 to S606 is performed using the number of candidates / selections set in step S611.

[0090] Then, when the residual falls below the convergence threshold, the control unit 26 outputs the latest design parameter x as the optimal design parameter (step S612: output step).

[0091] In the modified example 5 described above, similar to the embodiment, a three-dimensional model of the nozzle is created from the design parameters, and shape parameters representing the predicted fiber shape (cross-sectional shape) are extracted from the results of analyzing the material flow using the three-dimensional model. The optimal design parameters are then set based on the residuals with the shape parameters extracted immediately before. According to this modified example 5, by combining CFD simulation to analyze fluid flow, parameterization of analysis results by image analysis, and generation of design parameters by Bayesian optimization, appropriate conditions can be obtained in a short time when searching for design conditions for fibers.

[0092] Furthermore, according to this modified version 5, the number of candidates and selections for low-fidelity and high-fidelity CFD simulations executed during the iteration of the optimization process is reduced, thereby reducing the computational load associated with the iteration of CFD simulations and shortening the computation time compared to the process related to modified version 4.

[0093] (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 a molded product was made from elongated fibers, and design parameters for producing the molded product were explored. However, the present invention can also be applied to any molded product made by pouring and solidifying a liquid material. Examples include resin, metal, plastic, ceramic, rubber, and concrete. [Explanation of Symbols]

[0094] 1. Search System 2 Search device 3 Display device 4 Input devices 21 Three-Dimensional Model Creation Department 22 Analysis Department 23 Shape parameter extraction unit 24 Judgment section 25 Optimization Unit 26 Control Unit 27 Memory section

Claims

1. A computer search method for finding the optimal design parameters for obtaining a molded product having a predetermined material shape, Initial setup steps for setting design parameters and convergence thresholds for manufacturing molded products, A three-dimensional model creation step, in which a three-dimensional model is created based on the aforementioned design parameters, A simulation step in which a fluid simulation is performed using the three-dimensional model, A shape parameter extraction step in which shape parameters are extracted by image analysis of the simulation results, A convergence determination step that determines whether the difference between the shape parameter extracted in the shape parameter extraction step and the shape parameter / target value extracted in the extraction step performed immediately before is less than or equal to the convergence threshold, If the difference is determined to be greater than the convergence threshold, an update step is performed in which the design parameters are updated by an optimization algorithm using the set of all extracted shape parameters and design parameters. The iterative step involves performing the three-dimensional model creation step, the simulation step, the shape parameter extraction step, and the convergence determination step with respect to the design parameters updated by the update step, If it is determined that the difference is less than or equal to the convergence threshold, an output step is performed to output the design parameter to be determined as the optimal design parameter. A search method that includes this.

2. A simulation count determination step in the search process for the molded product, which determines whether it is the second or subsequent simulation, It further includes, The aforementioned convergence determination step is performed when it is determined that the simulation is running for the second time or later. The search method according to claim 1.

3. The update step updates the design parameters by Bayesian optimization. The search method according to claim 1 or 2.

4. The aforementioned update step is: A step to determine the number of updates, which determines whether the number of consecutive updates of the best value of the shape parameter is equal to or greater than a preset number of updates, If it is determined that the number of consecutive updates of the best value of the shape parameter is less than the set number, a modification step is taken to change the hyperparameter or acquisition function by Bayesian optimization. If it is determined that the number of consecutive updates of the best value of the shape parameter is equal to or greater than the set number, the design parameter is updated by Bayesian optimization using the hyperparameter or acquisition function used in the most recent update; if it is determined that the number of consecutive updates of the best value of the shape parameter is less than the set number, the design parameter is updated by Bayesian optimization using the hyperparameter or acquisition function changed in the change step; The search method according to claim 3, including the following:

5. The aforementioned update step is: A step to determine the number of updates, which determines whether the number of consecutive updates of the best value of the shape parameter is equal to or greater than a preset number of updates, If it is determined that the number of consecutive updates of the best value of the shape parameter is less than the set number, a modification step is taken to change the hyperparameter of the optimization algorithm. If it is determined that the number of consecutive updates of the best value of the shape parameter is equal to or greater than the set number, the design parameter is updated using the optimization algorithm with hyperparameters used in the previous update; if it is determined that the number of consecutive updates of the best value of the shape parameter is less than the set number, the design parameter is updated using the optimization algorithm with hyperparameters changed in the change step; The search method according to claim 1, including the following:

6. The aforementioned simulation step is A low-fidelity fluid simulation step, which involves performing a low-fidelity fluid simulation using the aforementioned three-dimensional model, A preliminary extraction step in which provisional shape parameters are extracted by image analysis of the low-fidelity fluid simulation results, A search and selection step involves searching for and selecting candidates from the extracted provisional shape parameters to be used in a high-fidelity fluid simulation, which has a higher computational load than the low-fidelity fluid simulation. A high-fidelity fluid simulation step in which the high-fidelity fluid simulation is performed using a three-dimensional model based on the design parameters selected in the exploration and selection step, The search method according to claim 3, including the following:

7. The aforementioned search and selection step involves selecting a predetermined number of design parameters, whose acquisition function values ​​based on the low-fidelity fluid simulation results are among the highest, as candidates for use in the high-fidelity simulation. The search method according to claim 6.

8. Before executing the aforementioned iterative step, a candidate / selection number update step is performed to reduce the number of candidates selected in the low-fidelity fluid simulation and the high-fidelity fluid simulation from the initial setting value. The search method according to claim 6, further comprising:

9. A search program that causes a computer to search for optimal design parameters for obtaining a molded product having a predetermined material shape, Initial setup steps for setting design parameters and convergence thresholds for manufacturing molded products, A three-dimensional model creation step, in which a three-dimensional model is created based on the aforementioned design parameters, A simulation step in which a fluid simulation is performed using the three-dimensional model, A shape parameter extraction step in which shape parameters are extracted by image analysis of the simulation results, A convergence determination step that determines whether the difference between the shape parameter extracted in the shape parameter extraction step and the shape parameter / target value extracted in the extraction step performed immediately before is less than or equal to the convergence threshold, If the difference is determined to be greater than the convergence threshold, an update step is performed in which the design parameters are updated by an optimization algorithm using the set of all extracted shape parameters and design parameters. The iterative step involves performing the three-dimensional model creation step, the simulation step, the shape parameter extraction step, and the convergence determination step with respect to the design parameters updated by the update step, If it is determined that the difference is less than or equal to the convergence threshold, an output step is performed to output the design parameter to be determined as the optimal design parameter. A search program that causes the aforementioned computer to execute.

10. A molded product manufactured using design parameters obtained by the search method described in claim 1.

11. A manufacturing method for producing a molded product using design parameters obtained by the search method described in claim 1.

Images