Resin molding analysis system, resin molding analysis method, program, and recording medium

Abstract

To provide a resin molding analysis system capable of easily and accurately performing analysis according to the shape of a molded article.SOLUTION: The resin molding analysis system 100 includes a server 20 and a client terminal 10 connected to the server 20 via a network 30. The client apparatus 10 performs machine-learning, generates a first learning model 12b for estimating an analysis parameter based on resin-molding condition information, and transmits the generated first learning model 12b to the server 20. The server 20 receives a plurality of first learning models 12b from the client terminals 10, integrates the plurality of received first learning models 12b, and generates a second learning model 12c for estimating an analysis parameter on the basis of resin-molding condition information.SELECTED DRAWING: Figure 1

Description

[Technical Field]

[0001] The present invention relates to a resin molding analysis system, a resin molding analysis method, a program, and a recording medium. [Background technology]

[0002] BACKGROUND ART Conventionally, an analyzer for a molded product is known (see, for example, Patent Document 1).

[0003] The above-mentioned Patent Document 1 discloses a molded product analysis device that performs analysis when a molded product is molded using a mold. The molded product analysis device in Patent Document 1 uses machine learning to learn the correlation between the shape characteristics of a molded product and defect information for the molded product, and derives a prediction formula that predicts defect information for the molded product from the shape characteristics of the molded product. Furthermore, the molded product analysis device predicts defect information from the shape characteristics of a new molded product based on the prediction formula predicted by machine learning. [Prior art documents] [Patent documents]

[0004] [Patent Document 1] Japanese Patent Publication No. 2023-049823 Summary of the Invention [Problem to be solved by the invention]

[0005] The molded product analysis device of Patent Document 1 uses machine learning to learn the correlation between the shape characteristics of a molded product and defect information for the molded product, and derives a prediction formula that predicts defect information for the molded product from the shape characteristics of the molded product. Therefore, while it is possible to predict defect information for a new molded product having shape characteristics similar to the shape characteristics of the molded product used in the machine learning based on the prediction formula predicted by machine learning, it is difficult to accurately predict defect information for a new molded product having shape characteristics that are not very similar to the shape of the molded product used in the machine learning. Therefore, in order to analyze molded products of various shapes, machine learning must be performed using molded products of many shapes, which poses a problem in that it is difficult to easily perform accurate analysis according to the shape of the molded product.

[0006] The present invention has been made to solve the above-mentioned problems, and one object of the present invention is to provide a resin molding analysis system, a resin molding analysis method, a program, and a recording medium that are capable of easily and accurately performing analysis according to the shape of the molded product. [Means for solving the problem]

[0007] In order to achieve the above-mentioned object, a resin molding analysis system according to a first aspect of the present invention comprises a server and a client terminal connected to the server via a network, wherein the client terminal is configured to perform a resin molding analysis that outputs characteristic information indicating the characteristics of a resin molded product molded in resin molding based on resin molding condition information and analysis parameters set when performing the simulation, and the client terminal performs machine learning to generate a first learning model that estimates analysis parameters based on the resin molding condition information and transmits the generated first learning model to the server, and the server receives multiple first learning models from the client terminal and integrates the received multiple first learning models to generate a second learning model that estimates analysis parameters based on the resin molding condition information.

[0008] In the resin molding analysis system according to the first aspect of the present invention, as described above, the server receives multiple first learning models from the client terminal and integrates the received multiple first learning models to generate a second learning model that estimates analysis parameters based on resin molding condition information. This allows the server to generate a second learning model that integrates multiple first learning models, making it possible to easily and accurately estimate analysis parameters for analyzing molded products of various shapes using the second learning model. As a result, analysis can be easily and accurately performed according to the shape of the molded product.

[0009] In the resin molding analysis system according to the first aspect, preferably, the client terminal transmits the first learning model to the server without transmitting the resin molding condition information used to generate the first learning model to the server. With this configuration, the resin molding condition information used to perform machine learning by each of the client terminals of the multiple users is not transmitted to the server, so that the resin molding condition information can be kept confidential from each of the multiple users. Furthermore, by transmitting the first learning models generated by machine learning by each of the client terminals of the multiple users to a common server, second learning models corresponding to the shapes of more molded products can be generated in the server.

[0010] In the resin molding analysis system according to the first aspect, the client terminal preferably generates a first learning model that estimates analysis parameters based on resin molding condition information and output values ​​based on actual measurement values ​​of properties of the corresponding resin molded product. With this configuration, it is possible to generate a first learning model that estimates analysis parameters for outputting output values ​​close to the actual measurement values ​​based on the resin molding condition information.

[0011] In the resin molding analysis system according to the first aspect, each of the first learning model and the second learning model preferably includes a learning network to which a plurality of nodes are connected, and when integrating the plurality of first learning models, the server integrates the weighting of calculations by the plurality of nodes to generate the second learning model. With this configuration, by adjusting and integrating the weighting of the nodes corresponding to each shape of the plurality of molded products, it is possible to easily generate second learning models corresponding to the shapes of many molded products.

[0012] In this case, the server preferably generates a second learning model by averaging the weights of the calculations of the multiple nodes when integrating the multiple first learning models. By averaging and integrating the weights of the nodes corresponding to the shapes of the multiple molded products, it is possible to more easily generate second learning models that correspond to the shapes of many molded products.

[0013] In the resin molding analysis system according to the first aspect, the client terminal preferably selects resin molding condition information and performs machine learning when generating the first learning model. With this configuration, machine learning for generating the first learning model can be performed efficiently.

[0014] In the resin molding analysis system according to the first aspect, the server preferably receives, in addition to the first learning model generated by the client terminal, resin molding condition information used when the first learning model was generated. This configuration allows the server to evaluate the reliability of the first learning model based on the resin molding condition information. As a result, it is possible to prevent the reliability of the second learning model generated by integrating a less reliable first learning model from being reduced.

[0015] A resin molding analysis method according to a second aspect of the present invention includes the steps of: performing a resin molding analysis to output characteristic information indicating the characteristics of a resin molded product molded in resin molding based on resin molding condition information and analysis parameters set when performing the simulation; performing machine learning to generate a first learning model that estimates analysis parameters based on the resin molding condition information; transmitting the generated first learning model to a server; and receiving multiple first learning models by the server and integrating the received multiple first learning models to generate a second learning model that estimates analysis parameters based on the resin molding condition information.

[0016] In the resin molding analysis method according to the second aspect of the present invention, as described above, the server receives multiple first learning models from the client terminal and integrates the received multiple first learning models to generate a second learning model that estimates analysis parameters based on resin molding condition information. This allows the server to generate a second learning model that integrates multiple first learning models, making it possible to easily and accurately estimate analysis parameters for analyzing molded products of various shapes using the second learning model. As a result, a resin molding analysis method can be provided that allows easy and accurate analysis according to the shape of the molded product.

[0017] A program according to a third aspect of the present invention causes a computer to execute the resin molding analysis method according to the second aspect.

[0018] A program according to a third aspect of the present invention allows a computer to execute the resin molding analysis method according to the second aspect, thereby enabling easy and accurate analysis in accordance with the shape of a molded product.

[0019] A storage medium according to a fourth aspect of the present invention has the program according to the third aspect recorded thereon and is computer-readable.

[0020] In a storage medium according to a fourth aspect of the present invention, by recording a program according to the third aspect, it is possible to provide a computer-readable storage medium that can easily and accurately perform analysis according to the shape of a molded product.

[0021] A resin molding analysis system according to a fifth aspect of the present invention includes a server and a client terminal that performs resin molding analysis and outputs characteristic information indicating the characteristics of a resin molded product molded in resin molding based on resin molding condition information and analysis parameters set when performing a simulation, wherein the client terminal performs machine learning and generates a first learning model that estimates analysis parameters based on the resin molding condition information, and the server acquires multiple first learning models generated in the client terminal and integrates the acquired multiple first learning models to generate a second learning model that estimates analysis parameters based on the resin molding condition information.

[0022] In the resin molding analysis system according to a fifth aspect of the present invention, as described above, the server acquires multiple first learning models generated by client terminals and integrates the acquired multiple first learning models to generate a second learning model that estimates analysis parameters based on resin molding condition information. This allows the server to generate a second learning model that integrates multiple first learning models, making it possible to easily and accurately estimate analysis parameters for analyzing molded products of various shapes using the second learning model. As a result, analysis can be easily and accurately performed according to the shape of the molded product.

[0023] In the resin molding analysis system according to the fifth aspect, the server is preferably configured to accept an input operation of the first learning model via a storage medium. With this configuration, even if the server and the client terminal are not connected via a network, the server can easily acquire the first learning model generated in the client terminal via the storage medium.

[0024] In the resin molding analysis system according to the fifth aspect, the server preferably acquires, in addition to the first learning model generated by the client terminal, resin molding condition information used when the first learning model was generated. This configuration allows the server to evaluate the reliability of the first learning model based on the resin molding condition information. As a result, it is possible to prevent the reliability of the second learning model generated by integrating a less reliable first learning model from being reduced.

[0025] A resin molding analysis method according to a sixth aspect of the present invention includes a step of performing a resin molding analysis that outputs characteristic information indicating the characteristics of a resin molded product molded in resin molding based on resin molding condition information and analysis parameters set when performing the simulation; a step of performing machine learning to generate a first learning model that estimates analysis parameters based on the resin molding condition information; and a step of acquiring multiple first learning models and integrating the acquired multiple first learning models to generate a second learning model that estimates analysis parameters based on the resin molding condition information.

[0026] A resin molding analysis method according to a sixth aspect of the present invention includes the steps of acquiring a plurality of first learning models, integrating the acquired plurality of first learning models, and generating a second learning model that estimates analysis parameters based on resin molding condition information, as described above. This allows the generation of a second learning model that integrates a plurality of first learning models, making it possible to easily and accurately estimate analysis parameters for analyzing molded products of various shapes using the second learning model. As a result, a resin molding analysis method can be provided that allows easy and accurate analysis according to the shape of the molded product. [Effects of the Invention]

[0027] According to the present invention, as described above, analysis can be easily performed with high accuracy according to the shape of the molded product. [Brief explanation of the drawings]

[0028] [Figure 1] 1 is a block diagram showing an example of a configuration for carrying out a resin molding analysis method according to a first embodiment. [Figure 2] 1A to 1C are diagrams for explaining a resin molding analysis method according to a first embodiment. [Figure 3] FIG. 2 is a diagram for explaining machine learning by a client terminal according to the first embodiment. [Figure 4] FIG. 2 is a diagram for explaining the integration of learning models by the server according to the first embodiment. [Figure 5] FIG. 3 is a diagram for explaining estimation of analysis parameters using a learning model according to the first embodiment. [Figure 6] 10 is a flowchart illustrating a procedure for updating a learning model according to the first embodiment. [Figure 7] FIG. 10 is a block diagram showing an example of a configuration for carrying out a resin molding analysis method according to a second embodiment. [Figure 8] FIG. 10 is a block diagram showing an example of a configuration for carrying out a resin molding analysis method according to a third embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0029] DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, specific embodiments of the present invention will be described with reference to the accompanying drawings.

[0030] [First embodiment] A resin molding analysis system 100 according to a first embodiment will be described with reference to FIGS.

[0031] The resin molding analysis system 100 according to the first embodiment performs flow analysis of a molded product and predicts the state of the molded product. For example, the resin molding analysis system 100 predicts the occurrence of defects in the molded product.

[0032] (Device configuration example) A resin molding analysis system 100 according to the first embodiment implements a resin molding analysis method. The resin molding analysis method can be implemented, for example, by an apparatus configuration such as that shown in Fig. 1. The resin molding analysis system 100 includes a plurality of client terminals 10 and a server 20. The plurality of client terminals 10 and the server 20 are connected via a network 30.

[0033] The client terminal 10 is configured to be able to execute the program 12a. The client terminal 10 is a computer. In addition, a part or all of the processing performed by causing the client terminal 10 to execute the program 12a may be performed by hardware such as a dedicated arithmetic circuit. The client terminal 10 is located in the facility of a user who performs resin molding analysis. In addition, each of the multiple client terminals 10 is used by a different user.

[0034] The server 20 is configured to be able to execute the program 22a. The server 20 is a computer. Some or all of the processing performed by causing the server 20 to execute the program 22a may be performed by hardware such as a dedicated arithmetic circuit. The server 20 is located in the facility of an administrator who manages the resin molding analysis system 100.

[0035] 1, the client terminal 10 includes one or more processors 11, such as a central processing unit (CPU), and a storage unit 12 including a read-only memory (ROM), a random access memory (RAM), and a storage device, such as a hard disk drive or a semiconductor storage device.

[0036] The client terminal 10 can perform resin molding analysis by causing the processor 11 to execute the program 12a stored in the storage unit 12. The program 12a may be read from a recording medium 17 or may be provided from an external server or the like via a transmission path such as a network such as the Internet or a LAN (Local Area Network). The recording medium 17 is a computer-readable recording medium such as an optical disk, a magnetic disk, or a nonvolatile semiconductor memory, and has the program 12a recorded therein.

[0037] In addition to the program 12a, various analysis data used for resin molding analysis are stored in the storage unit 12. The analysis data includes resin molding information including resin molding condition information and characteristics of the resin molded product, parameters for simulating the state of the resin during molding and the details of the characteristics of the molded product, product category information, shape feature values, a group of optimal parameters, error rates, numerical data used for analysis, data on analysis conditions, etc.

[0038] The memory unit 12 also stores learning models (first learning model 12b, second learning model 12c) used to perform resin molding analysis. The first learning model 12b is a learning model created in each of the multiple client terminals 10. The second learning model 12c is a learning model generated by the server 20 based on the multiple first learning models 12b.

[0039] The client terminal 10 also includes a display unit 13 such as a liquid crystal display device, an input unit 14 consisting of input devices such as a keyboard and a mouse, a communication unit 15 for communicating via the network 30, and a reading unit 16 for reading the program 12a and various data from a recording medium 17. The reading unit 16 is a reader device or the like according to the type of recording medium 17. The user can input data for analysis conditions using the input unit 14. The analysis data may be read from a recording medium created by the user, or may be created by the user on an external server or the like and obtained from the external server via a transmission path.

[0040] The server 20 includes one or more processors 21, each of which includes a central processing unit (CPU) or the like, and a storage unit 22 including a read-only memory (ROM), a random access memory (RAM), and a storage device, such as a hard disk drive or a semiconductor storage device.

[0041] The server 20 can perform processing by causing the processor 21 to execute the program 22a stored in the storage unit 22. The program 22a may be read from the recording medium 17 or may be provided from an external server or the like via a transmission path such as a network such as the Internet or a LAN (Local Area Network).

[0042] The memory unit 22 stores a plurality of first learning models 12b transmitted from each of a plurality of client terminals 10, and a second learning model 12c generated based on the plurality of first learning models 12b.

[0043] (Resin molding analysis method) Next, a resin molding analysis method will be described. In the first embodiment, the resin molding analysis method performs a resin flow analysis of a molded product of an arbitrary shape.

[0044] In the resin molding analysis method, as shown in FIG. 2, a client terminal 10 performs a simulation based on resin molding condition information and analysis parameters, and outputs characteristic information of the resin molded product. Specifically, in the resin molding analysis method, the shape of the molded product, material data, and molding conditions are input to the client terminal 10 as resin molding condition information for a molded product of an arbitrary shape. In addition, analysis parameters for performing the resin molding analysis (simulation) are input to the client terminal 10 by the user. In the resin molding analysis method, analysis mesh information in which the molded product is divided into multiple analysis meshes (small elements), material data, molding conditions, and analysis parameters are input, and injection molding analysis is performed based on the input information. Then, as analysis results, information (characteristic information of the resin molded product) such as the resin filling pattern, resin pressure, temperature, resin orientation, physical properties, warpage, roundness, filling pattern, geometric tolerance, and predicted locations of appearance defects (welds, sink marks, flow marks, burns, silver streaks) is output.

[0045] For example, the information entered to perform an analysis includes product category information (application, field), analysis (calculation) mesh information (element type, number of elements, number of nodes, division conditions, element quality), resin data (resin manufacturer, grade name, base resin, latent heat, solidification temperature, density, specific heat, thermal conductivity, melt viscosity, PVT data, elastic modulus, Poisson's ratio, linear expansion coefficient, molding shrinkage rate, mechanical strength, reinforcement properties, reinforcement content, viscoelastic properties (Prony series, shift factor), optical properties (stress optical coefficient, photoelastic coefficient, refractive index, molecular structure, gelation reaction rate, curing reaction heat), molding conditions (time, This includes information such as filling rate, upper pressure limit, screw position, screw speed, flow rate, metering position, resin temperature, mold temperature, VP switching timing, packing pressure, packing time, in-mold cooling conditions, cycle time), mold conditions (nozzle position, gate position, parting surface, number of gates, runner layout, cooling circuit, ejector pin arrangement), boundary conditions (nozzle flow rate and pressure, heat transfer coefficient, ambient temperature, ambient humidity), molding machine information (molding machine manufacturer name, molding machine model number, maximum injection speed, maximum injection pressure, maximum holding pressure, screw diameter, maximum mold clamping force, maximum injection capacity), and rigidity information of the molded product.

[0046] Analysis parameters are set when a simulation is performed. Specifically, the analysis parameters can be set by the user who performs the analysis. However, when the parameters are diverse and complex, the analysis results will vary depending on the skill of the user who sets the parameter values.

[0047] However, the mathematical models used in simulations do not fully reflect the actual phenomenon, and modeling errors occur. In actual phenomena, values ​​are not always constant, and many variations are involved, such as the effect of different resin lots on physical properties and the heat transfer coefficient inside the mold. However, because it is difficult to grasp these facts precisely, ideal conditions are often assumed and constant values ​​are set over time and space, and these effects also become a source of error.

[0048] For example, consider a case where the measured warpage of an injection-molded product does not match the predicted value (analysis result). To investigate the cause, the factors behind the difference between the analysis and the actual measurement must first be identified. Analysis allows data to be saved during calculation, allowing it to be retrieved and verified later. On the other hand, with an actual molded product, there is no historical information remaining from the time the molten resin is poured into the mold, to the time it is cooled and removed, making it difficult to compare the two. As a result, it is usually difficult to identify the cause of the difference between the measured value and the analysis value.

[0049] The analysis software is equipped with analytical parameters that can be used to correct the analysis and adjust the accuracy. By adjusting the analytical parameter values, it is possible to make corrections to the analysis and estimate parameters that match the actual measured values. Various methods have been studied to efficiently estimate the optimal parameters. For example, injection molding simulation can be combined with parametric optimization to calculate injection molding simulations with gradually changing parameter values, and the result with the smallest error can be adopted.

[0050] In addition, it is known that injection molding simulations have strong nonlinearities between control variables (input values) and target variables (output values). When nonlinearities are strong, machine learning techniques such as multilayer neural networks can be used as a parameter estimation method.

[0051] Here, in the first embodiment, the client terminal 10 performs machine learning to generate a first learning model 12b that estimates analysis parameters based on resin molding condition information. The client terminal 10 also transmits the generated first learning model 12b to the server 20. The machine learning is performed, for example, by a full-layer convolutional network.

[0052] 4, the server 20 receives a plurality of first learning models 12b from the client terminal 10, integrates the received plurality of first learning models 12b, and generates a second learning model 12c that estimates analysis parameters based on resin molding condition information. The server 20 then transmits the generated second learning model 12c to each of the plurality of client terminals 10.

[0053] The client terminal 10 transmits the first learning model 12b to the server 20 without transmitting the resin molding condition information used when generating the first learning model 12b to the server 20. In other words, the client terminal 10 does not transmit the actual data used when generating the first learning model 12b to the server 20. Therefore, the actual data of the user using the client terminal 10 that generated the first learning model 12b is not transmitted to the server 20.

[0054] 3, the client terminal 10 generates a first learning model 12b that estimates analysis parameters based on resin molding condition information and output values ​​based on actual measurement values ​​of the properties of the corresponding resin molded product. Note that the actual measurement values ​​may be, for example, measurement values ​​of a molded product similar to the injection molded product, empirical values, estimated values, etc.

[0055] For example, when generating the first learning model 12b, machine learning is performed to optimize analysis parameters based on one type of analysis result and the corresponding measurement result (actual measurement value). Furthermore, machine learning may be performed to optimize analysis parameters so that multiple types of analysis results match each corresponding measurement result (actual measurement value). For example, in the flow analysis, a mold cooling analysis is performed, a filling analysis is performed, and a pressure-holding cooling analysis, shrinkage warpage analysis, and fiber orientation analysis are performed in parallel. In this case, analysis parameters may be calculated simultaneously so that multiple analysis results match the measurement result.

[0056] When generating the first learning model 12b, the client terminal 10 selects resin molding condition information and performs machine learning. For example, the client terminal 10 selects resin molding condition information that is effective for machine learning and uses it for machine learning.

[0057] Regarding dataset selection, a large amount of model data is required for accurate machine learning. Furthermore, regular updating of training data is necessary. Specifically, this can be expressed as (annotation cost) = (cost per data point) × (total number of data points). In other words, data selection is important for reducing the total amount of data. Randomly selecting data reduces the learning effectiveness of the learning model. Active learning is a technique for selecting data with high learning effectiveness from unlabeled data. Because of its high learning effectiveness, it is possible to achieve high accuracy even with a small amount of data. For example, data selection can be performed using pool-based sampling, which trains an initial model from a small amount of labeled data and selects the data considered to be most effective for label learning. Data selection can also be performed using stream-based selective sampling, which selects valuable data from a stream, labels it, and discards the rest. Data selection can also be performed using membership query synthesis, which generates data useful for model training.

[0058] Furthermore, as shown in FIG. 5, the client terminal 10 uses a learning model (the first learning model 12b or the second learning model 12c) to estimate analysis parameters based on resin molding condition information.

[0059] By estimating analysis parameters using the first learning model 12b, it is possible to set analysis parameters for performing accurate analysis under similar conditions (conditions with similar shape, molding conditions, and resin used) when performing a new analysis.

[0060] For example, in a resin molding analysis method, analysis is performed by setting at least one of the resin flow characteristics, resin physical properties, and warpage parameters as analysis parameters. Resin flow characteristics include, for example, MFR, MVR, melt viscosity, flow length, melting point, glass transition point, and flow length measurement results using a spiral flow mold. Resin physical properties include, for example, density, specific heat, thermal conductivity, elastic modulus, Poisson's ratio, linear expansion coefficient, PVT characteristics, molding shrinkage, and reinforcement properties. Warpage parameters include, for example, shrinkage distortion correction, rigidity correction, mold constraint effect, filler orientation correction, nozzle boundary condition (temperature and pressure) correction, stress relaxation correction, heat transfer coefficient adjustment, and shrinkage onset detection parameters. The analysis parameters may be the resin flow characteristics, resin physical properties, and warpage parameters themselves, or may be based on the resin flow characteristics, resin physical properties, and warpage parameters. Furthermore, the analysis parameters are not limited to those exemplified above.

[0061] As shown in FIG. 3, each of the first learning model 12b and the second learning model 12c includes a learning network N to which multiple nodes N1 to N8 are connected. Note that the connection structure of each node and the number of nodes are not limited to those shown in FIG. 3. FIG. 3 shows the learning network N schematically; in reality, the number of nodes is greater than eight, and the connection structure of each node is also complex. Furthermore, the number of nodes and the connection structure of each node in the learning networks N of the multiple first learning models 12b and the second learning model 12c are the same. This makes it possible to easily generate the second learning model 12c by integrating multiple first learning models 12b.

[0062] When integrating multiple first learning models 12b, server 20 integrates the weighting of calculations by multiple nodes N1 to N8 to generate second learning model 12c. For example, when integrating multiple first learning models 12b, server 20 averages and integrates the weighting of calculations by multiple nodes N1 to N8 to generate second learning model 12c.

[0063] (Learning model update process) Referring to FIG. 6, the update process of the learning model used when acquiring analytical parameters for performing flow analysis will be described.

[0064] In step S1 of FIG. 6, the server 20 initializes a global learning model. The global learning model is, for example, a general AI learning model. In step S2, the server 20 distributes the global learning model to each of the multiple client terminals 10. Note that data communication between the server 20 and each client terminal 10 is subject to privacy protection measures such as encryption.

[0065] In step S3, each of the multiple client terminals 10 performs machine learning from the global learning model, and a local first learning model 12b is generated. In machine learning in the client terminal 10, it is preferable to incorporate new data each time machine learning is performed. Note that the client terminal 10 may perform machine learning using the same data each time machine learning is performed. In step S4, the multiple first learning models 12b generated by each of the multiple client terminals 10 are aggregated in the server 20. Note that privacy protection measures such as encryption are implemented in data communication between the server 20 and each client terminal 10.

[0066] In step S5, the server 20 integrates multiple first learning models 12b to generate a second learning model 12c. For example, the second learning model 12c is generated by averaging the nodes of the AI ​​model. In step S6, the server 20 evaluates the performance of the second learning model 12c. The generated second learning model 12c is distributed to each of the multiple client terminals 10, and the performance of the second learning model 12c is evaluated in each of the multiple client terminals 10. If the performance of the second learning model 12c achieves the predetermined evaluation, the learning model update process ends. If the performance of the second learning model 12c does not achieve the predetermined evaluation, the processes of steps S3 to S6 are repeated.

[0067] In other words, the server 20 and the client terminal 10 perform a simulation using analysis parameters estimated using the second learning model 12c, generate a first learning model 12b, and integrate the first learning model 12b to generate a second learning model 12c, repeatedly until the output characteristic information reaches a predetermined accuracy target value.

[0068] (Effects of the first embodiment) The effects of the first embodiment will be described.

[0069] In the first embodiment, as described above, the server 20 receives multiple first learning models 12b from the client terminal 10, integrates the received multiple first learning models 12b, and generates a second learning model 12c that estimates analysis parameters based on resin molding condition information. This allows the server 20 to generate a second learning model 12c that integrates multiple first learning models 12b, so that analysis parameters for analyzing molded products of various shapes can be easily and accurately estimated using the second learning model 12c. As a result, analysis can be easily and accurately performed according to the shape of the molded product.

[0070] Furthermore, in the first embodiment, as described above, the client terminal 10 transmits the first learning model 12b to the server 20 without transmitting the resin molding condition information used to generate the first learning model 12b. As a result, the resin molding condition information used to perform machine learning by each of the client terminals 10 of the multiple users is not transmitted to the server 20, so the resin molding condition information can be kept confidential from each of the multiple users. Furthermore, by transmitting the first learning models 12b generated by machine learning by each of the client terminals 10 of the multiple users to the common server 20, second learning models 12c corresponding to the shapes of more molded products can be generated in the server 20.

[0071] Furthermore, in the first embodiment, as described above, the client terminal 10 generates the first learning model 12b that estimates analysis parameters based on resin molding condition information and output values ​​based on actual measurement values ​​of the properties of the corresponding resin molded product. This makes it possible to generate the first learning model 12b that estimates analysis parameters for outputting output values ​​close to the actual measurement values ​​based on the resin molding condition information.

[0072] Furthermore, in the first embodiment, as described above, each of the first learning model 12b and the second learning model 12c includes a learning network N to which multiple nodes N1 to N8 are connected, and when integrating multiple first learning models 12b, the server 20 integrates the weighting of calculations by the multiple nodes N1 to N8 to generate the second learning model 12c. This makes it possible to easily generate a second learning model 12c that corresponds to the shapes of many molded products by adjusting and integrating the weighting of the nodes N1 to N8 that correspond to each of the shapes of multiple molded products.

[0073] Furthermore, in the first embodiment, as described above, when integrating multiple first learning models 12b, the server 20 averages and integrates the weightings of the calculations by multiple nodes N1 to N8 to generate the second learning model 12c. This makes it easier to generate second learning models 12c that correspond to the shapes of many molded products by averaging and integrating the weightings of the nodes N1 to N8 that correspond to the shapes of each of multiple molded products.

[0074] Furthermore, in the first embodiment, as described above, when generating the first learning model 12b, the client terminal 10 selects resin molding condition information and performs machine learning. This allows for efficient machine learning to generate the first learning model 12b.

[0075] [Second embodiment] Next, a resin molding analysis system 200 according to a second embodiment will be described with reference to FIG.

[0076] Unlike the first embodiment, in which the client terminal does not transmit to the server the resin molding condition information used when the first learning model was generated, in the second embodiment, at least one client terminal is configured to transmit to the server the resin molding condition information used when the first learning model was generated. The other configurations of the second embodiment are the same as those of the first embodiment. The same reference numerals are used for the same configurations as those of the first embodiment, and descriptions thereof will be omitted.

[0077] 7, a resin molding analysis system 200 in the second embodiment includes a plurality of client terminals 10 and 210, and a server 220. The client terminal 210 and the server 220 are computers, similar to the first embodiment.

[0078] The client terminal 210 is configured to transmit resin molding condition information used when the first learning model 12b was generated to the server 220. Specifically, in the second embodiment, the server 220 is configured to receive (acquire) the resin molding condition information used when the first learning model 12b was generated, in addition to the first learning model 12b generated in the client terminal 210. Here, at least one of the multiple client terminals is the client terminal 210.

[0079] As shown in FIG. 7 , the memory unit 22 of the server 220 stores the first learning model 12b generated by the client terminal 210 and the resin molding condition information corresponding to the first learning model 12b. This allows a user who can access the server 220 to evaluate the reliability of the first learning model 12b corresponding to the resin molding condition information based on the resin molding condition information stored in the memory unit 22. If the user determines that the reliability of the first learning model 12b does not meet a predetermined level as a result of the evaluation, the user can take measures to prevent the reliability of the generated second learning model 12c from being reduced, such as excluding the first learning model 12b determined to be unreliable from the integration targets for generating the second learning model 12c, or improving the first learning model 12b determined to be unreliable using the resin molding condition information to improve its reliability. The evaluation of the reliability of the first learning model 12b based on the resin molding condition information may be performed by the user as described above, or by the processor 21 of the server 220.

[0080] (Effects of the second embodiment) Next, the effects of the second embodiment will be described.

[0081] In the second embodiment, as described above, the server 220 receives (acquires) the resin molding condition information used when the first learning model 12b was generated in addition to the first learning model 12b generated in the client terminal 210. This allows the server 220 to evaluate the reliability of the first learning model 12b based on the resin molding condition information. As a result, it is possible to prevent the reliability of the second learning model 12c generated by integrating a first learning model 12b with low reliability from being reduced. Note that other effects of the second embodiment are the same as those of the first embodiment.

[0082] [Third embodiment] Next, a resin molding analysis system 300 according to a third embodiment will be described with reference to FIG.

[0083] Unlike the first embodiment in which all client terminals are connected to the server via a network, in the third embodiment, at least one of the multiple client terminals is not connected to the server via a network. The other configurations of the third embodiment are the same as those of the first embodiment. The same reference numerals are used for the same configurations as those of the first embodiment, and descriptions thereof will be omitted.

[0084] As shown in Fig. 8, a resin molding analysis system 300 in the third embodiment includes a plurality of client terminals 10 and 310, and a server 320. The client terminals 310 and 320 are computers, similar to the first embodiment. As shown in Fig. 8, the client terminal 10 is connected to the server 320 via a network 30. On the other hand, the client terminal 310 is not connected to the server 320 via the network 30.

[0085] As shown in FIG. 8, the server 320 is configured to acquire the first learning model 12b generated in the client terminal 310, for example, via a USB memory 330. In the third embodiment, the server 320 is configured to accept an input operation of the first learning model 12b via the USB memory 330. Specifically, the processor 21 of the server 320 is configured to accept an input operation of the first learning model 12b via the USB memory 330. As shown in FIG. 8, the server 320 includes a reading unit 324, which is a reader device corresponding to the USB memory 330, and is configured to accept an input operation of the first learning model 12b via the reading unit 324. The USB memory 330 is an example of a "storage medium" in the claims.

[0086] As described above, in the third embodiment, the server 320 acquires multiple first learning models 12b generated in the client terminals 10 and 310, and integrates the acquired multiple first learning models 12b to generate a second learning model 12c that estimates analysis parameters based on resin molding condition information.

[0087] In addition, in combination with the second embodiment, the server 320 may acquire the first learning model 12b generated in the client terminal 310 and the resin molding condition information used when the first learning model 12b was generated via the USB memory 330.

[0088] (Effects of the third embodiment) Next, the effects of the third embodiment will be described.

[0089] In the third embodiment, as described above, the server 320 is configured to accept input operations for the first learning model 12b via the USB memory 330. As a result, even if the server 320 and the client terminal 310 are not connected via a network, the server 320 can easily acquire the first learning model 12b generated in the client terminal 310 via the USB memory 330. Note that other effects of the third embodiment are similar to those of the first embodiment.

[0090] (Variation) The embodiments disclosed herein should be considered to be illustrative and not restrictive in all respects. The scope of the present invention is defined by the claims rather than the above description of the embodiments, and further includes all modifications (variations) within the meaning and scope of the claims.

[0091] For example, in the first to third embodiments, the resin molding analysis system includes a plurality of client terminals, but the present invention is not limited to this. In the present invention, the resin molding analysis system may include a single client terminal. In this case, a second learning model may be generated by integrating multiple first learning models generated by the single client terminal using a server.

[0092] Furthermore, in the first embodiment, for convenience of explanation, the processing operation of a computer is described using a flow-driven flowchart in which processing is performed sequentially according to a processing flow, but the present invention is not limited to this. In the present invention, the processing operation of a computer may be performed by event-driven processing in which processing is performed on an event-by-event basis. In this case, the processing may be performed completely event-driven, or may be performed in a combination of event-driven and flow-driven.

[0093] In the first to third embodiments, a resin molding analysis system performs a simulation and outputs characteristics of a resin molded product that indicate defects in the resin molded product. However, the present invention is not limited to this. In the present invention, estimated costs may be output as characteristics of the resin molded product. That is, a client terminal may be configured to perform a resin molding analysis that outputs estimated costs for a mold for resin molding as characteristic information based on resin molding condition information and analysis parameters. For example, in a globally operating company, an automatic estimate for components, which are in-house know-how, may be output by the resin molding analysis system. This allows client terminals in multiple factories to estimate the required material costs, processing costs, etc., based on product shapes (e.g., 3D shape data). First, a global model is created, for example, in a factory in Japan. A global model (machine learning model) is created using information that remains in the domestic factory. In this case, a machine learning model is generated using the product shape (featurized), product material costs, processing method, and processing accuracy as input values, and the relationship between the estimated costs as output values. This global model is then sent to factories in Europe and Asia, where it is combined with the product information held by each factory to create a local AI model. The parameters of this local model are sent to domestic factories, which then update the global model, which is then sent back to each client. This process is repeated several times until a satisfactory level of accuracy is achieved.

[0094] Alternatively, for example, a resin molding analysis may be performed as follows to obtain an estimated cost for a mold for resin molding. First, CAD data is input into the client terminal that performs the resin molding analysis, and the client terminal analyzes the product shape based on the input data. In this case, the client terminal checks the product's complexity, size, and shape features (presence or absence of undercuts and ribs). Next, the material to be used for the product is selected, and the mold is designed based on the shape analysis results and material selection. Design details include mold division, gate position, runner system, cooling system, ejection system, etc. Next, the manufacturing method (e.g., CNC machining, EDM) is selected. Once the mold design and manufacturing method have been decided, a cost estimate is made. The cost takes into account material costs, processing costs, design costs, and other related expenses. Finally, a quote is prepared.

[0095] In this case, for example, first, data held by each factory is collected and a local model (machine learning model) is created for each factory. An AI model is also created from shape analysis, materials, and design details. The parameters of the trained local model are sent to the server. In this case, no data is sent, and only the learning model information is shared. The global model is updated based on the parameters received by the server. This global model will integrate the knowledge of each factory. The updated global model is distributed to each factory, and the local models are updated. The updated global model is then distributed to each factory, and the local models are updated. By repeating this process, the accuracy of the model is improved. One application of using the improved accuracy learning model in each factory is, for example, shape analysis and material selection when designing a new shape, allowing estimates to be made based on past knowledge, making it possible to quickly propose the product to the customer.

[0096] In addition, in the above second embodiment, an example was shown in which multiple client terminals include a client terminal 10 that sends the first learning model 12b to the server 220 without sending resin molding condition information, and a client terminal 210 that sends both the resin molding condition information and the first learning model 12b to the server 220, but the present invention is not limited to this. In the present invention, all client terminals may send both the resin molding condition information and the first learning model to the server.

[0097] In the third embodiment, a USB memory is used as the "storage medium" in the claims, but the present invention is not limited to this. In the present invention, a hard disk, a CD, a DVD, an SD card, etc. may be used as the "storage medium" in the claims.

[0098] In addition, in the third embodiment, an example was shown in which some of the client terminals 310 among the multiple client terminals were not connected to the server 320 via the network 30, but the present invention is not limited to this. In the present invention, all client terminals do not need to be connected to the server via the network.

[0099] In addition, in the above third embodiment, an example was shown in which a USB memory 330 is used to share with the server 320 the first learning model 12b generated in a client terminal 310 that is not connected to the server 320 via the network 30, but the present invention is not limited to this. In the present invention, a storage medium may be used to share with the server the first learning model and resin molding condition information generated in a client terminal that is connected to the server via a network. [Explanation of symbols]

[0100] 10, 210, 310 Client terminal (computer) 12a Program 12b First Learning Model 12c Second Learning Model 17 Recording Media 20, 220, 320 Server (computer) 22a Program 30 Network 100, 200, 300 Resin molding analysis system 330 USB memory (storage medium) N Learning Network N1, N2, N3, N4, N5, N6, N7, N8 nodes

Claims

1. A server; a client terminal connected to the server via a network, the client terminal is configured to perform a resin molding analysis that outputs characteristic information indicating characteristics of a resin molded product molded in resin molding, based on resin molding condition information and analysis parameters set when performing a simulation; the client terminal performs machine learning to generate a first learning model that estimates the analysis parameters based on the resin molding condition information, and transmits the generated first learning model to the server; The server receives multiple first learning models from the client terminal, integrates the received multiple first learning models, and generates a second learning model that estimates the analysis parameters based on the resin molding condition information.

2. The resin molding analysis system of claim 1, wherein the client terminal transmits the first learning model to the server without transmitting the resin molding condition information used when generating the first learning model to the server.

3. 2. The resin molding analysis system of claim 1, wherein the client terminal generates the first learning model that estimates the analysis parameters based on the resin molding condition information and output values ​​based on actual measured values ​​of the characteristics of the corresponding resin molded product.

4. each of the first learning model and the second learning model includes a learning network to which a plurality of nodes are connected; The resin molding analysis system according to claim 1 , wherein the server, when integrating the plurality of first learning models, integrates the weighting of calculations by the plurality of nodes to generate the second learning model.

5. The resin molding analysis system of claim 4, wherein the server, when integrating multiple first learning models, averages and integrates the weighting of calculations by multiple nodes to generate the second learning model.

6. The resin molding analysis system according to claim 1 , wherein the client terminal selects the resin molding condition information and performs machine learning when generating the first learning model.

7. The resin molding analysis system of claim 1, wherein the server and the client terminal perform a simulation using the analysis parameters estimated using the second learning model, generate the first learning model, and integrate the first learning model to generate the second learning model until the output characteristic information reaches a predetermined accuracy target value.

8. 2. The resin molding analysis system according to claim 1, wherein the client terminal is configured to perform a resin molding analysis that outputs an estimated cost of a mold for resin molding as the characteristic information based on the resin molding condition information and the analysis parameters.

9. The resin molding analysis system described in claim 1, wherein the server receives, in addition to the first learning model generated at the client terminal, the resin molding condition information used when the first learning model was generated.

10. a step of performing a resin molding analysis to output characteristic information indicating characteristics of a resin molded product molded in resin molding based on resin molding condition information and analysis parameters set when performing the simulation; performing machine learning to generate a first learning model that estimates the analysis parameters based on the resin molding condition information; transmitting the generated first learning model to a server; A resin molding analysis method comprising a step of receiving multiple first learning models by the server, integrating the received multiple first learning models, and generating a second learning model that estimates the analysis parameters based on the resin molding condition information.

11. A program that causes a computer to execute the resin molding analysis method according to claim 10.

12. A recording medium on which the program according to claim 11 is recorded and which is readable by the computer.

13. A server; a client terminal that performs resin molding analysis and outputs characteristic information indicating the characteristics of a resin molded product molded in resin molding based on resin molding condition information and analysis parameters that are set when performing a simulation, the client terminal performs machine learning to generate a first learning model that estimates the analysis parameters based on the resin molding condition information; The server acquires multiple first learning models generated at the client terminal, integrates the acquired multiple first learning models, and generates a second learning model that estimates the analysis parameters based on the resin molding condition information.

14. The resin molding analysis system according to claim 13 , wherein the server is configured to accept an input operation for the first learning model via a storage medium.

15. The resin molding analysis system described in claim 13, wherein the server acquires, in addition to the first learning model generated in the client terminal, the resin molding condition information used when the first learning model was generated.

16. a step of performing a resin molding analysis to output characteristic information indicating characteristics of a resin molded product molded in resin molding based on resin molding condition information and analysis parameters set when performing the simulation; performing machine learning to generate a first learning model that estimates the analysis parameters based on the resin molding condition information; A resin molding analysis method comprising a step of acquiring a plurality of the first learning models, integrating the acquired plurality of first learning models, and generating a second learning model that estimates the analysis parameters based on the resin molding condition information.

Images