Predictive model generation system, program, and predictive model generation method

Abstract

This invention provides a predictive model generation system, method, and program that predicts the characteristic values ​​of resin products based on their manufacturing conditions. [Solution] A program executed by a computer that causes the computer to function as a predictive model generation system, wherein the predictive model generation system 10 includes a search unit that searches for manufacturing conditions that give feature values ​​that satisfy predetermined conditions, using a predictive model that predicts the feature values ​​of a resin product generated on CAE (Computer Aided Engineering) using the manufacturing conditions of the resin product, and a model learning unit that updates the predictive model using data including the manufacturing conditions obtained by the search and the feature values ​​of the resin product simulated on CAE using the manufacturing conditions.

Description

Technical Field

[0001] The present invention relates to a prediction model generation system, a program, and a prediction model generation method.

Background Art

[0002] Patent Document 1 describes "a program for causing a computer to execute a process of calculating a stiffness matrix including a plurality of stiffness values corresponding to the plurality of edges and a force vector including a plurality of force values corresponding to the plurality of nodes based on mesh data including a plurality of nodes and a plurality of edges connecting the plurality of nodes and boundary condition data indicating a force applied to an object represented by the mesh data, generating feature data from the stiffness matrix and the force vector, and inferring a plurality of displacement amounts corresponding to the plurality of nodes by performing a convolution operation on the feature data according to the connection relationship of the plurality of nodes shown by the mesh data." (Claim 1). Patent Document 2 describes "equipment configuration acquisition means for acquiring current equipment configuration information indicating the current component configuration of a machine tool and the current control program of the machine tool, operation history acquisition means for acquiring operation history information indicating the operation history of the machine tool, accuracy achievement acquisition means for acquiring accuracy achievement information indicating the actual machining accuracy of the machine tool, characteristic identification means for identifying accuracy-related characteristics that are characteristics affecting the machining accuracy of the machine tool based on the current equipment configuration information, the operation history information, and the accuracy achievement information, user requirement acquisition means for acquiring a user's requirement regarding machining accuracy, product information acquisition means for acquiring product information indicating at least one of the parts that can be used for the machine tool and the control programs that can be used for the machine tool, equipment configuration determination means for determining a proposed equipment configuration that satisfies the user's requirement and in which at least one of the current component configuration and the current control program in the current equipment configuration is changed based on the current equipment configuration information, the accuracy-related characteristics, and the product information, and information output means for outputting proposed equipment-related information that is information regarding the proposed equipment configuration." (Claim 1). [Prior Art Documents] [Patent] [Patent Document 1] Japanese Unexamined Patent Publication No. 2023-9904 [Patent Document 2] WO2023 / 153446 [Overview of the project]

[0003] In a first embodiment of the present invention, a predictive model generation system comprising a search unit and a model learning unit is provided. The search unit searches for manufacturing conditions that satisfy predetermined conditions and provide feature values, using a predictive model that predicts the feature values ​​of a resin product generated on CAE (Computer Aided Engineering) using the manufacturing conditions of the resin product. The model learning unit updates the predictive model using the manufacturing conditions obtained through the search and data including the feature values ​​of the resin product simulated on CAE using the manufacturing conditions.

[0004] In the above, the search unit may include a prediction unit that inputs manufacturing conditions into a prediction model and outputs feature values.

[0005] In the above, the search unit may have a condition generation unit that generates manufacturing conditions used for the search.

[0006] In the above, the search unit may include an evaluation unit that inputs manufacturing conditions into a prediction model and evaluates the obtained feature values.

[0007] In the above, the evaluation unit may output an evaluation value obtained by evaluating the feature value, and the prediction unit may search for manufacturing conditions in which an evaluation value equal to or greater than a threshold is predicted.

[0008] In the above, the evaluation unit may search for and output multiple manufacturing conditions that give an evaluation value equal to or greater than a threshold.

[0009] In the above, the condition generation unit may generate new manufacturing conditions by applying a genetic algorithm to multiple manufacturing conditions that give an evaluation value above a threshold.

[0010] In the above, the evaluation unit may search for and output multiple manufacturing conditions that give a Pareto solution in a feature space defined by the numerical values ​​of the feature values.

[0011] In the above, the search unit may have an unexplored identification unit that identifies unexplored regions in the manufacturing condition space defined by numerical values ​​of manufacturing conditions, where the condition generation unit has not generated any manufacturing conditions. The condition generation unit may generate manufacturing conditions that are included in the unexplored regions.

[0012] In the above, the evaluation unit may perform an additional search using the prediction model updated by the model learning unit to further search for manufacturing conditions in which the prediction unit predicts an evaluation value equal to or greater than a threshold. The model learning unit may perform an additional update to update the prediction model using the manufacturing conditions obtained through the additional search and the pairs of feature values ​​of the resin product simulated by CAE using those manufacturing conditions.

[0013] In the above, the prediction model generation system may repeatedly perform additional exploration and additional update processes. The evaluation unit may terminate the iteration in accordance with the fact that, within the manufacturing condition space defined by the numerical values ​​of the manufacturing conditions, the movement of manufacturing conditions for which the prediction unit predicts an evaluation value above a threshold falls below a threshold.

[0014] In the above, the manufacturing conditions may include at least one piece of information regarding the materials of the resin product and the processing conditions of the resin product.

[0015] In the above, the characteristic value may include at least one of the following: performance and / or physical properties of the resin product, numerical values ​​relating to its shape, and numerical values ​​relating to the conditions necessary for its manufacture.

[0016] In the above, the resin product may be an injection-molded resin product. CAE may include simulating the injection molding process of the resin product.

[0017] In the above, the resin product may include an anisotropic material having anisotropy. CAE may include simulating the orientation of the anisotropic material.

[0018] In the above, the prediction model may input the orientation information regarding the orientation of the anisotropic material obtained by executing CAE using the manufacturing conditions, and predict the characteristic values.

[0019] In the above, the prediction model may input the manufacturing conditions, and predict the characteristic values of the resin product generated by considering the orientation of the anisotropic material on CAE based on the manufacturing conditions.

[0020] In the above, the anisotropic material may be a fiber material.

[0021] In the above, the prediction model generation system may include a display unit. The display unit may compress the manufacturing condition space defined by the numerical values of the manufacturing conditions into a two-dimensional plane, and display the manufacturing conditions searched by the search unit on the two-dimensional plane.

[0022] In the above, the display unit may display an unexplored plane area on the two-dimensional plane where the condition generation unit has not generated the manufacturing conditions.

[0023] In the above, the model learning unit may update the prediction model using a pair of the specified manufacturing conditions included in the unexplored plane area specified by the user and the characteristic values of the resin product simulated by CAE with the specified manufacturing conditions.

[0024] In the above, the prediction model generation system may include a CAE execution unit. The CAE execution unit may simulate the manufacturing of the resin product under the manufacturing conditions obtained by the search by CAE.

[0025] In a second aspect of the present invention, there is provided a program that is executed by a computer and causes the computer to function as the above-described prediction model generation system.

[0026] In a third aspect of the present invention, a method for generating a prediction model including a search stage and a model learning stage is provided. In the search stage, based on the manufacturing conditions of a resin product, using a prediction model that predicts characteristic values of the resin product generated on CAE (Computer Aided Engineering) using the manufacturing conditions, manufacturing conditions that give predetermined characteristic values are searched for. In the model learning stage, the prediction model is updated using the manufacturing conditions obtained by the search and the pairs of the characteristic values of the resin product simulated by CAE under the manufacturing conditions.

[0027] In a fourth aspect of the present invention, a prediction system including a prediction unit is provided. The prediction unit predicts, using a prediction model, characteristic values of a resin product in a state where an anisotropic material is oriented on CAE (Computer Aided Engineering) based on characteristic values in an unoriented state of a resin material including an orientable anisotropic material and orientation information regarding the orientation of the anisotropic material included in the resin material.

[0028] In the above, the characteristic values may be variables affected by the orientation of the anisotropic material. The orientation information may include the orientation direction of the anisotropic material specified by a flow analysis.

[0029] In the above, the characteristic values may at least include any one of the stress distribution, the coefficient of thermal expansion distribution, and the deformation of the resin product.

[0030] In a fifth aspect of the present invention, a program executed by a computer to cause the computer to function as the above prediction system is provided.

[0031] In a fifth aspect of the present invention, a prediction method including a prediction stage is provided, in which, based on characteristic values in an unoriented state of a resin material including an orientable anisotropic material and orientation information regarding the orientation of the anisotropic material included in the resin material, characteristic values of a resin product in a state where the anisotropic material is oriented generated on CAE (Computer Aided Engineering) are predicted using a prediction model.

[0032] It should be noted that the above summary of the invention does not enumerate all of its features. Furthermore, subcombinations of these features may also constitute an invention. [Brief explanation of the drawing]

[0033] [Figure 1] The configuration of the prediction model generation system 10 according to this embodiment is shown. [Figure 2] This embodiment shows the relationship between the prediction model and CAE. [Figure 3] The flowchart for the prediction model generation method according to this embodiment is shown. [Figure 4] Figure 3 shows an example of the subflow of S300. [Figure 5] A graph illustrating an example of the relationship between manufacturing conditions and evaluation values ​​is shown on a two-dimensional plane. [Figure 6] An example of initial manufacturing conditions and an initial prediction model is shown. [Figure 7] An example of manufacturing conditions and a predictive model after one search is shown. [Figure 8] An example of manufacturing conditions and a predictive model after three exploration iterations is shown. [Figure 9] An example of manufacturing conditions and a predictive model after further exploration is shown. [Figure 10] Figure 9 shows an example of manufacturing conditions and a predictive model after further exploration. [Figure 11] An example of the subflow of S300 in a modified example is shown. [Figure 12] An example of displaying unexplored regions is shown. [Figure 13] Examples of a computer 2200 in which multiple aspects of the present invention may be embodied in whole or in part are shown. [Figure 14] An example of anisotropy exhibited by anisotropic materials is shown. [Figure 15] An example of stress acting on a small cubic region is shown. [Figure 16] The relationship between the prediction model and CAE in a second modified example of this embodiment is shown. [Figure 17]The relationship between the prediction model and CAE in a third modified example of this embodiment is shown. [Modes for carrying out the invention]

[0034] The present invention will be described below through embodiments, but these embodiments are not intended to limit the scope of the claims. Furthermore, not all combinations of features described in the embodiments are necessarily essential to the solution of the invention.

[0035] Figure 1 shows the configuration of the predictive model generation system 10 according to this embodiment. The predictive model generation system 10 generates a predictive model, such as a surrogate model, that predicts the feature values ​​of a resin product generated on CAE (Computer Aided Engineering) using the manufacturing conditions of the resin product.

[0036] The predictive model generation system 10 comprises a CAE execution unit 110, a model learning unit 120, a search unit 130, and a display unit 140. The predictive model generation system 10 may also include modules with other functions as needed.

[0037] The prediction model generation system 10 may be a computer such as a PC (personal computer), tablet computer, smartphone, workstation, server computer, or general-purpose computer, and may also be a computer system in which multiple computers are connected.

[0038] Alternatively, the predictive model generation system 10 may be a dedicated computer designed for predictive model generation processing, or dedicated hardware implemented by dedicated circuits. The predictive model generation system 10 may be implemented by a single device (computer), or by multiple devices with assigned roles. In the predictive model generation system 10, although not specifically described below, memory / hard disks, etc., are provided, and information necessary for processing is stored as appropriate, and information is transmitted between each processing module such as the CAE execution unit 110 and the model learning unit 120.

[0039] The CAE execution unit 110 simulates the manufacturing of a resin product under given manufacturing conditions using CAE (Computer-Aided Engineering). By executing CAE, the CAE execution unit 110 can reproduce a resin product manufactured under arbitrary manufacturing conditions on a computer without actually performing prototyping or experiments. The CAE execution unit 110 outputs feature values ​​that represent the characteristics of the resin product generated by CAE. The CAE execution unit 110 may be implemented using an existing CAE program.

[0040] The CAE execution unit 110 may perform CAE on predetermined manufacturing conditions and / or manufacturing conditions specified by the user. For example, the CAE execution unit 110 may perform CAE on multiple manufacturing conditions specified as initial manufacturing conditions. Furthermore, the CAE execution unit 110 may perform CAE on manufacturing conditions recommended by the search unit 130 (also called "recommended manufacturing conditions").

[0041] The model learning unit 120 learns and / or updates a prediction model. The prediction model predicts the feature values ​​of a resin product generated on CAE, using at least partially the manufacturing conditions of the resin product. For example, the prediction model may output feature values ​​by directly inputting the manufacturing conditions, or it may output desired feature values ​​by inputting parameters derived from the manufacturing conditions.

[0042] The model learning unit 120 learns a predictive model using training data. The model learning unit 120 may obtain training data from an external database or the like. The training data includes multiple pairs of explanatory variables and target variables. For example, such pairs include manufacturing conditions and feature values ​​of resin products simulated by the CAE execution unit 110 using CAE based on those manufacturing conditions.

[0043] The model learning unit 120 may update the prediction model. For example, the model learning unit 120 may update the prediction model using the results of CAE execution performed by the CAE execution unit 110 under recommended manufacturing conditions as training data. The model learning unit 120 may only update the prediction model. In this case, the prediction model generation system 10 may initially use a pre-provided prediction model.

[0044] The search unit 130 uses a prediction model to search for manufacturing conditions that provide feature values ​​that satisfy predetermined conditions. For example, the search unit 130 searches for manufacturing conditions such as the feature value being above a threshold, falling within a predetermined range, and / or the evaluation value obtained by evaluating the feature value being above a threshold. For example, the search unit 130 includes a condition generation unit 132, a prediction unit 134, an evaluation unit 136, and an unsearched condition identification unit 138.

[0045] The condition generation unit 132 generates manufacturing conditions to be used for the search. The condition generation unit 132 may generate manufacturing conditions for when the CAE execution unit 110 executes CAE. For example, the condition generation unit 132 may generate manufacturing conditions to be used in the initial stages of the search. Details of the processing of the condition generation unit 132 will be described later.

[0046] The prediction unit 134 inputs the manufacturing conditions generated by the condition generation unit 132 into the prediction model learned by the model learning unit 120 and outputs feature values. This makes it possible to predict the feature values ​​of resin products generated by CAE under specific manufacturing conditions without performing CAE.

[0047] The evaluation unit 136 evaluates the feature values ​​(also called "predicted feature values") obtained by the prediction unit 134 by inputting manufacturing conditions into the prediction model. For example, the evaluation unit 136 may output an evaluation value obtained by evaluating the feature values.

[0048] Based on the evaluation, the evaluation unit 136 causes the condition generation unit 132 to generate new manufacturing conditions using a search method. For example, the evaluation unit 136 may cause the condition generation unit 132 to generate manufacturing conditions such that the evaluation provides feature values ​​that satisfy predetermined conditions.

[0049] The evaluation unit 136 may determine whether the evaluation of the feature values ​​satisfies predetermined conditions. The evaluation unit 136 provides the CAE execution unit 110 with manufacturing conditions that give feature values ​​that satisfy the predetermined conditions as recommended manufacturing conditions.

[0050] The unexplored area identification unit 138 obtains the manufacturing conditions generated by the condition generation unit 132 and identifies unexplored areas where the exploration of manufacturing conditions is insufficient. For example, the unexplored area identification unit 138 identifies unexplored areas in the manufacturing condition space defined by the numerical values ​​of the manufacturing conditions, where the condition generation unit 132 has not yet generated any manufacturing conditions. By providing the unexplored areas to the condition generation unit 132, the unexplored area identification unit 138 assists the condition generation unit 132 in generating manufacturing conditions included in the unexplored areas.

[0051] The display unit 140 displays the processing results from the prediction model generation system 10. For example, the display unit 140 displays the CAE execution results from the CAE execution unit 110, the search results from the search unit 130, and / or the prediction results from the prediction unit 134.

[0052] The predictive model generation system 10 does not necessarily have to include a CAE execution unit 110. In this case, instead of using the CAE execution unit 110, the predictive model generation system 10 may acquire feature values ​​obtained by performing CAE from an external source, and / or acquire feature values ​​obtained by actually prototyping and measuring resin products.

[0053] As described above, the prediction model generation system 10 of this embodiment searches for recommended manufacturing conditions, performs CAE under those recommended manufacturing conditions, and updates the prediction model based on the results. This makes it possible to obtain a highly accurate prediction model in the vicinity of the recommended manufacturing conditions without excessive CAE execution. This reduces the number of CAE executions while maintaining the accuracy of the prediction model. Alternatively, a prediction system equipped with at least a prediction unit 134 may be provided instead of the prediction model generation system 10. In this case, the learning and generation of the prediction model may be performed outside the prediction system.

[0054] Figure 2 shows the relationship between the prediction model and CAE according to this embodiment. When the CAE execution unit 110 etc. executes CAE 210 based on the manufacturing conditions of the resin product, the resin product manufactured on the computer under those manufacturing conditions is reproduced, and the characteristic values ​​of the resin product can be obtained on the computer.

[0055] By training a model with pairs of manufacturing conditions and feature values ​​obtained from CAE (Computer-Aided Engineering), a predictive model 220 (also called a "surrogate model") can be generated that predicts feature values ​​from manufacturing conditions. The feature values ​​output by the predictive model 220 from the manufacturing conditions are only predictions and may differ from the feature values ​​actually obtained by performing CAE under those manufacturing conditions. In general, the more training data containing CAE results there is, the higher the accuracy of the predictive model 220 can be.

[0056] However, executing CAE requires significant computing, memory, and / or time resources. Therefore, it is desirable to minimize the execution of CAE as much as possible without significantly sacrificing the accuracy of the predictive model. According to this embodiment, CAE execution can be concentrated on manufacturing conditions that are highly valued and important, thereby enabling the construction of a highly accurate surrogate model while suppressing the consumption of computing, memory, and / or time resources.

[0057] Figure 3 shows the flow of the method for generating a predictive model according to this embodiment. The predictive model generation system 10 generates a predictive model that predicts the characteristics of resin products by executing each of the processes S100 to S700, for example. The order of the processes S100 to S700 may be changed, and some processes may be omitted.

[0058] The type of resin product is not particularly limited. The resin product may be a product containing resin manufactured using various materials and / or manufacturing methods. In particular, this embodiment is suitably used for injection-molded resin products manufactured by methods in which processing conditions, etc., affect the characteristics and quality of the product, such as injection molding.

[0059] First, in S100, the predictive model generation system 10 inputs initial data. The initial data includes information about the resin product and its manufacturing (also called "manufacturing information") necessary for CAE execution, excluding manufacturing conditions, i.e., fixed conditions within the manufacturing information that are not the subject of the search (also called "fixed conditions"). The manufacturing information may be selected from the following: dimensions, shape, mass of the resin product, resin material (e.g., substance name, product name, chemical formula, composition, viscosity, specific heat and / or other physical properties), name of the manufacturing equipment, structure of the manufacturing equipment (e.g., mold shape and gate position), processing capacity of the manufacturing equipment (e.g., machinable dimensions and / or weight), operating conditions of the manufacturing equipment (e.g., operating conditions such as voltage and processing speed), processing conditions (e.g., mold temperature, resin temperature, processing temperature and / or processing pressure), boundary conditions of the resin product (e.g., load conditions and constraint conditions), etc.

[0060] For example, the initial data may include the dimensions and / or shape of the resin product. As an example, the initial data may include mesh shape setting conditions, which are the conditions for dividing the shape of the resin product into a mesh on the CAE (for example, conditions for dividing the shape of the resin product into a collection of many meshes).

[0061] The initial data may include information about the prediction model. For example, it may include information about the type of prediction model and / or its hyperparameters.

[0062] The initial data may include a specification of the search range for manufacturing conditions. The condition generation unit 132 generates manufacturing conditions so that they fall within the specified search range.

[0063] The initial data may include the termination conditions for the search. For example, the initial data may include the target value for the evaluation value.

[0064] Initial data may be input by the user to the predictive model generation system 10. Alternatively, the initial data may be data already stored in the predictive model generation system 10 or a database, etc.

[0065] Next, in S150, the condition generation unit 132 generates manufacturing conditions (also called "initial manufacturing conditions") to be used in the initial stages of the search. The manufacturing conditions may be those manufacturing information that are the subject of the search, for example, manufacturing information excluding fixed conditions such as shape (e.g., mesh shape setting conditions). As an example, the manufacturing conditions may include at least one piece of information such as the material of the resin product, the processing conditions of the resin product, and the structure of the manufacturing apparatus.

[0066] The condition generation unit 132 may generate multiple initial manufacturing conditions using a known algorithm so that the manufacturing conditions are distributed without bias. For example, the condition generation unit 132 may generate initial manufacturing conditions by randomly sampling multiple coordinates in a manufacturing condition space defined by multiple manufacturing condition variables. For example, the condition generation unit 132 may generate initial manufacturing conditions using experimental design, D-optimization criterion experimental design, full condition search, or known sampling methods such as LHS.

[0067] The condition generation unit 132 may generate initial manufacturing conditions within the search range. The initial manufacturing conditions may be specified directly by the user. The initial manufacturing conditions may also be those that are pre-stored in the prediction model generation system 10.

[0068] Next, in S200, the CAE execution unit 110 simulates the manufacturing of the resin product under initial manufacturing conditions using CAE (Computer-Aided Engineering). For example, the CAE execution unit 110 may simulate the injection molding process of the resin product using CAE. As a result, the CAE execution unit 110 outputs feature values ​​that represent the characteristics of the resin product generated by CAE under the initial manufacturing conditions.

[0069] Feature values ​​are information that describes the characteristics of a resin product and may be represented by one or more variables. For example, feature values ​​may include at least one of the following: performance and / or physical properties of the resin product, numerical values ​​related to its shape, and numerical values ​​related to the conditions required for its manufacture. Feature values ​​may also include text information other than numerical values.

[0070] The CAE execution unit 110 may provide the model learning unit 120 with a set of pairs of initial manufacturing conditions and feature values ​​obtained by performing CAE under those initial manufacturing conditions as initial training data.

[0071] Thereafter, the prediction model generation system 10 may repeat the loop from S250 to S700 until the termination condition is met. In this loop, the execution of CAE using the manufacturing conditions obtained from the search using the prediction model and the updating of the prediction model using the training data obtained from said CAE may be repeated alternately.

[0072] Next, in S250, the model learning unit 120 learns a prediction model. In the first S250, the model learning unit 120 generates a prediction model based on the initial training data provided by the CAE execution unit 110. The model learning unit 120 learns a prediction model that takes the manufacturing conditions included in the initial training data as explanatory variables and outputs the feature values ​​corresponding to those manufacturing conditions as the target variable.

[0073] The model learning unit 120 may use a known model as the prediction model. The type of prediction model is not particularly limited, but an extrapolable model can be suitably used. For example, the model learning unit 120 may learn the prediction model using a deep learning model such as a neural network, and / or a regression model such as ridge regression or support vector regression (SVR).

[0074] The model learning unit 120 may optimize the hyperparameters of the prediction model in each or some of the S250 processes. For example, the model learning unit 120 may optimize the R 2 CV You can use this as an indicator to automatically adjust the hyperparameters of the predictive model through cross-validation.

[0075] For example, when using SVR, cross-validation can be efficiently performed in the following way. First, the model learning unit 120 selects the Gaussian kernel parameter γ such that the variance of the Gram matrix is ​​maximized, determines C to be the larger of (the absolute value of (mean + standard deviation × 3 of γ) or (mean - standard deviation × 3 of γ)), and optimizes only ε by cross-validation. Next, the model learning unit 120 uses the optimized value for ε, selects γ such that the variance of the Gram matrix is ​​maximized, and then optimizes C by cross-validation. Finally, the model learning unit 120 fixes ε and C to the optimized values ​​and optimizes γ by cross-validation.

[0076] In subsequent S250 iterations, the model learning unit 120 updates the prediction model using additional training data obtained in the preceding S700. For example, the model learning unit 120 may update the prediction model by adding the training data from the most recent S700 to the training data accumulated so far and training a new prediction model. Alternatively, the model learning unit 120 may update the prediction model by performing transfer learning on the latest prediction model using the training data from the most recent S700.

[0077] Next, in S300, the prediction model generation system 10 generates and displays the prediction results from the prediction model. For example, the display unit 140 displays the prediction results from the prediction model by projecting the predicted values ​​from the prediction model generated in the preceding S250 onto the manufacturing condition space or the like.

[0078] Figure 4 shows an example of the subflow of S300 in Figure 3. The prediction model generation system 10 may execute S300 by executing S310 to S50 in Figure 4. Some of S310 to S350 may be omitted, and the order of processing may be changed as necessary.

[0079] In S310, the display unit 140 generates uniformly distributed points in the manufacturing condition space defined by the numerical values ​​of the manufacturing conditions. The manufacturing condition space may be a multidimensional space with the same number of coordinates as the variables included in the manufacturing conditions. The display unit 140 may generate uniformly distributed points such that the manufacturing condition space coincides with the search range of the manufacturing conditions, or is a narrower range than the search range.

[0080] The display unit 140 may generate a number of points uniformly distributed within the coordinate space using a known method. The display unit 140 may generate a predetermined number of points, or a number of points specified by the user.

[0081] In S320, the display unit 140 generates dummy data. For example, the display unit 140 may generate multiple points as dummy data in any way in the manufacturing condition space. The display unit 140 may select points to be dummy data from the uniformly distributed points generated in S310. For example, the display unit 140 may generate one or more points as dummy data randomly selected in the manufacturing condition space.

[0082] In S330, the display unit 140 compresses the manufacturing condition space into a two-dimensional plane and displays the manufacturing conditions on which CAE has been performed on the two-dimensional plane. Based on the manufacturing conditions corresponding to the uniformly distributed points generated in S310, the manufacturing conditions corresponding to the dummy data generated in S320, and the manufacturing conditions on which CAE was performed in S250, the display unit 140 displays the manufacturing conditions on which CAE has been performed up to that point on the two-dimensional plane. The manufacturing conditions on which CAE has been performed up to that point (also called "CAE-executed manufacturing conditions") may be the manufacturing conditions on which CAE was performed in S200 and the preceding S500.

[0083] The display unit 140 may use existing dimensionality reduction methods to compress the manufacturing condition space into a two-dimensional plane and display the CAE-executed manufacturing conditions on the two-dimensional plane. For example, the display unit 140 defines the entire manufacturing condition space based on the uniformly distributed points generated in S310, and compresses the manufacturing condition space into a two-dimensional plane so that the combined manufacturing conditions of the dummy data and the CAE-executed manufacturing conditions in S250 coincide with the entire two-dimensional plane to be displayed. The display unit 140 may then plot only the CAE-executed manufacturing conditions in S250 on the two-dimensional plane.

[0084] The display unit 140 may use, for example, UMAP (Uniform Manifold Approximation and Projection) or t-SNE (t-distributed Stochastic Neighbor Embedding) as a dimensionality reduction method.

[0085] In S340, the display unit 140 acquires predicted feature values ​​output by the prediction model for each point on the two-dimensional plane. For example, the display unit 140 may supply manufacturing conditions corresponding to any multiple points on the two-dimensional plane to the prediction unit 134, have the prediction unit 134 generate predicted values ​​using the prediction model generated in the most recent S250 based on those manufacturing conditions, and acquire these values.

[0086] In S350, the display unit 140 may display the predicted values ​​obtained in S340 on a two-dimensional plane, overlaid with the plot. For example, the display unit 140 may display each coordinate of a manufacturing condition on the two-dimensional plane in a different color depending on the predicted value predicted for that manufacturing condition. For example, the display unit 140 may color the two-dimensional plane such that the color of the point corresponding to the manufacturing condition becomes lighter as the predicted value for that manufacturing condition improves.

[0087] The above example describes a manufacturing condition space compressed into two dimensions, but it is not limited to this. For example, the prediction model generation system 10 may display the manufacturing condition space compressed into three dimensions.

[0088] In S400, following S300, the search unit 130 searches for manufacturing conditions. The search unit 130 may use the prediction model generated in the most recent S250 to search for manufacturing conditions.

[0089] The evaluation unit 136 searches for manufacturing conditions that the prediction unit 134 predicts will result in an evaluation value equal to or greater than a threshold. For example, the evaluation unit 136 causes the condition generation unit 132 to generate new candidate manufacturing conditions (also called "candidate manufacturing conditions") such that the evaluation value is equal to or greater than a threshold, and supplies these to the prediction unit 134. The prediction unit 134 inputs the candidate manufacturing conditions into the prediction model generated in the most recent S250, obtains the corresponding feature values, and outputs them to the evaluation unit 136. The condition generation unit 132 may generate candidate manufacturing conditions within a pre-specified search range.

[0090] The evaluation unit 136 generates evaluation values ​​by evaluating the feature values ​​obtained from the candidate manufacturing conditions. For example, the evaluation unit 136 may generate evaluation values ​​by inputting the feature values ​​into an evaluation function. The evaluation function may be a function that generates higher evaluation values ​​the more desirable each feature value is. The evaluation unit 136 may use a known function such as a linear function as the evaluation function.

[0091] The evaluation unit 136 provides evaluation value information to the condition generation unit 132, and causes the condition generation unit 132 to generate new manufacturing conditions that improve the evaluation value using a known search method such as the gradient method. The evaluation unit 136 may also use the results of S340 to identify manufacturing conditions that result in a good evaluation value.

[0092] The evaluation unit 136 may search for multiple manufacturing conditions that yield an evaluation value equal to or greater than a threshold. That is, the evaluation unit 136 may instruct the condition generation unit 132 to continue generating manufacturing conditions until a predetermined number of manufacturing conditions that yield an evaluation value equal to or greater than a threshold are reached.

[0093] The condition generation unit 132 may generate new manufacturing conditions based on multiple manufacturing conditions that give an evaluation value of a threshold or higher. For example, if the condition generation unit 132 finds one manufacturing condition that gives an evaluation value of a threshold or higher, it may generate one or more other manufacturing conditions in the vicinity of that manufacturing condition (for example, those whose distance in the manufacturing condition space is within a threshold). For example, the condition generation unit 132 may generate new manufacturing conditions by applying a genetic algorithm to multiple manufacturing conditions that give an evaluation value of a threshold or higher. This makes it possible to efficiently generate manufacturing conditions with high evaluation values. The evaluation unit 136 supplies the manufacturing conditions obtained as a result of the search to the CAE execution unit 110 as recommended manufacturing conditions.

[0094] When multiple evaluation values ​​are used, the evaluation unit 136 may search for and output multiple manufacturing conditions that give a Pareto solution in the evaluation value space defined by the numerical values ​​of the evaluation values. For example, the evaluation unit 136 may generate a Pareto solution curve or surface from multiple neighboring manufacturing conditions that form a Pareto solution, and generate new manufacturing conditions by randomly selecting manufacturing conditions on the Pareto solution curve or surface.

[0095] The evaluation unit 136 may search for manufacturing conditions using the feature values ​​themselves rather than evaluation values. For example, the evaluation unit 136 may search for manufacturing conditions that give feature values ​​that are above a threshold, below a threshold, and / or within a predetermined range. When multiple feature values ​​are used, the evaluation unit 136 may search for and output multiple manufacturing conditions that give a Pareto solution in the feature space defined by the numerical values ​​of the feature values.

[0096] In the second and subsequent S400, the evaluation unit 136 uses the prediction model updated by the model learning unit 120 in the most recent S250 to perform an additional search to further search for manufacturing conditions in which the prediction unit 134 predicts an evaluation value above a threshold. In the next S250, the model learning unit 120 performs an additional update to update the prediction model using the manufacturing conditions obtained through the additional search and the pair of feature values ​​of the resin product simulated by CAE in S500, which will be described later, based on those manufacturing conditions. The prediction model generation system 10 may repeatedly perform the additional search and additional update processes.

[0097] In S500, the CAE execution unit 110 simulates the manufacturing of a resin product using CAE based on the recommended manufacturing conditions obtained from the search in the most recent step, S400. The CAE execution unit 110 may perform the CAE in the same manner as in S200. The CAE execution unit 110 may provide the model learning unit 120 with the recommended manufacturing conditions and multiple pairs of feature values ​​obtained by performing CAE under those recommended manufacturing conditions as training data for updating.

[0098] In S600, the prediction model generation system 10 determines whether the termination conditions for the search are met. Any conditions can be set as termination conditions. For example, the evaluation unit 136 may determine that the termination conditions are met when: the change in the prediction model and / or recommended manufacturing conditions in the loop from S250 to S700 becomes sufficiently small; the number of CAE-executed manufacturing conditions reaches a certain number; a predetermined number or all of the feature values ​​obtained by the CAE executed in S500 achieve the target; the evaluation value obtained by the feature values ​​obtained by the CAE executed in S500 achieves the target; the evaluation value obtained by the feature values ​​obtained by the CAE executed in S500 does not update to the highest value for a certain period of time and / or a certain number of consecutive times; a certain number of loop processes have been executed; and / or a certain period of time has elapsed.

[0099] For example, the evaluation unit 136 terminates the loop when, within the loop iteration, the movement of manufacturing conditions in the manufacturing condition space for which the prediction unit 134 predicts an evaluation value of a threshold or higher falls below the threshold. For example, the evaluation unit 136 may determine that the termination condition is met when it has executed the loop processing a set number of times or more, and the movement of manufacturing conditions for which an evaluation value of a threshold or higher is predicted falls below the threshold.

[0100] In this case, the evaluation unit 136 terminates the iteration if the manufacturing conditions explored in the most recent S400 have traveled less than or equal to a threshold within the manufacturing condition space compared to the manufacturing conditions explored in the previous S400 (i.e., there has been little movement within the manufacturing condition space). This is because it has determined that there is little need for further exploration when the newly obtained manufacturing conditions have hardly changed as a result of the exploration.

[0101] As an example, the evaluation unit 136 may terminate the iteration when the amount of change in the parameters constituting the prediction model (e.g., regression coefficients, etc.) before and after the execution of S250 (e.g., change in the sum of all parameters, change in the mean, and / or change in the vector composed of all parameters, etc.) falls below a threshold. This is because it has been determined that continuing the search will not result in a significant change in the prediction model.

[0102] If the iteration is terminated, the prediction model generation system 10 may terminate the process. Alternatively, the prediction model generation system 10 may execute the process in S300 again and display the final prediction result, CAE-executed manufacturing conditions, and / or the prediction model, etc. If the iteration is not terminated, the prediction model generation system 10 proceeds to process S700.

[0103] In S700, the CAE execution unit 110 adds training data based on the results of S500. For example, the CAE execution unit 110 adds a pair of manufacturing conditions for which CAE was executed in the most recent S500, and the feature values ​​of the resin product simulated by CAE under those manufacturing conditions, to the training data. Next, the predictive model generation system 10 proceeds to S250 and repeats the loop processing.

[0104] In the next step, S250, the model learning unit 120 can update the prediction model using the added training data. This updates the prediction model using data including the manufacturing conditions obtained through the search in S400, and the feature values ​​of the resin product simulated by CAE in S500 using those manufacturing conditions.

[0105] The following examples of displays by the display unit 140 in S300 will be explained with reference to Figures 5 to 11, etc.

[0106] Figure 5 shows a graph illustrating an example of the relationship between manufacturing conditions and evaluation values ​​on a two-dimensional plane. The vertical and horizontal axes of the graph represent the axes in a plane that compresses the manufacturing condition space into two dimensions. The intensity of the colors in the graph indicates the magnitude of the evaluation values ​​obtained by evaluating the characteristic values ​​corresponding to the manufacturing conditions using an evaluation function; lighter colors indicate higher evaluation values.

[0107] The graph in Figure 5 is an example of true data representing the relationship between manufacturing conditions and evaluation values. The predictive model generation system 10 is assumed to train a predictive model so that it can predict the true data in Figure 5.

[0108] Figure 6 shows an example of initial manufacturing conditions and an initial prediction model. The graph in Figure 6 is a graph in which the manufacturing condition space is transformed into a two-dimensional plane, similar to Figure 5. The six X values ​​contained in region 610 correspond to the initial manufacturing conditions generated in S150. The intensity of the colors in the graph corresponds to the prediction results of the prediction model trained based only on the initial manufacturing conditions.

[0109] In other words, the graph in Figure 6 shows the results of running CAE in S200 under initial manufacturing conditions, using the results as training data to train a prediction model in S250, predicting the feature values ​​of each point on a two-dimensional plane, and then representing the evaluation values ​​obtained by transforming those feature values ​​with an evaluation function, shown in varying shades. As shown in the figure, at this stage, the prediction model is generated using only six initial data points, so the accuracy of the prediction is rough, and the degree of agreement with the true data in Figure 5 is still low.

[0110] The display unit 140 may display an image like that shown in Figure 6 during the processing of S100 to S700. For example, the display unit 140 may display an image like that shown in Figure 6 during the first S330 to S350.

[0111] Figure 7 shows an example of manufacturing conditions and a prediction model after one search. The display unit 140 displays the manufacturing conditions searched by the search unit 130 on a two-dimensional plane. In Figure 7, in addition to the initial manufacturing conditions of region 610, the manufacturing conditions of region 710 are included. The X included in region 710 corresponds to the recommended manufacturing conditions obtained in the first search in S400. The intensity of the colors in the graph corresponds to the prediction results of the prediction model learned based on the initial manufacturing conditions and the added manufacturing conditions. Region 710 may be the region containing the manufacturing conditions that are expected to yield the best evaluation value as a result of searching using the prediction model.

[0112] In other words, the graph in Figure 7 shows the results of running CAE in S200 and S500 using the initial manufacturing conditions and the recommended manufacturing conditions obtained from the first search. The prediction model, which was then trained again in S250 using the results as training data, was made to predict the feature values ​​of each point on a two-dimensional plane. The evaluation values ​​obtained by transforming these feature values ​​using an evaluation function are represented by varying shades of gray. The graph in Figure 7 is more complex than the graph in Figure 6 and is closer to the true data in Figure 5.

[0113] The display unit 140 may display an image like that shown in Figure 7 during the processing of S100 to S700. For example, during the first search in S400, the first execution of S500 to S700, and S330 to S350 after the second execution of S250, the display unit 140 may display the image shown in Figure 7.

[0114] Figure 8 shows an example of manufacturing conditions and a predictive model after three explorations. In Figure 8, in addition to regions 610 and 710, manufacturing conditions for regions 810 and 820 are included. The X values ​​included in regions 810 and 820 correspond to the recommended manufacturing conditions obtained in the second and third S400 explorations. The intensity of the colors in the graph corresponds to the prediction results of the predictive model learned based on the initial and added manufacturing conditions. Regions 810 and 820 may be regions containing manufacturing conditions that are expected to yield the best evaluation value based on the results of exploration using the predictive model at each of the two and three explorations.

[0115] In other words, the graph in Figure 8 shows the results of CAE execution in S200 and S500 using the initial manufacturing conditions and the recommended manufacturing conditions obtained from the 1st to 3rd searches. The prediction model, which was then trained again in S250 using the results as training data, was made to predict the feature values ​​of each point on a two-dimensional plane, and the evaluation values ​​obtained by transforming these feature values ​​with an evaluation function are represented by varying shades of gray. The graph in Figure 8 is more complex than the graphs in Figures 6 and 7 and is closer to the true data in Figure 5.

[0116] The display unit 140 may display an image like that shown in Figure 7 during the processing of S100 to S700. For example, during the first search in S400, the first execution of S500 to S700, and S330 to S350 after the second execution of S250, the display unit 140 may display the image shown in Figure 7.

[0117] In this way, the predictive model generation system adds the results of CAE execution of recommended manufacturing conditions to the training data and updates the predictive model. If the recommended manufacturing conditions are those with high ratings, the CAE will be performed intensively on those high-rated manufacturing conditions, thus maintaining a high prediction accuracy in the manufacturing condition areas that receive particularly high ratings.

[0118] Figure 9 shows an example of manufacturing conditions and a predictive model after further exploration. Figure 10 shows an example of manufacturing conditions and a predictive model after further exploration following Figure 9. The predictive model generation system 10 continues the exploration after the state in Figure 8 and obtains the manufacturing conditions included in region 910 in Figure 9 as recommended manufacturing conditions. Figure 10 shows that after the state in Figure 9, the exploration in S400 is performed again and manufacturing conditions included in region 1010 are obtained.

[0119] As shown in the figure, the manufacturing conditions included in region 910 and the manufacturing conditions included in region 1010 are close together. This indicates that as the exploration progresses and the update range of the prediction model decreases, the changes in the recommended manufacturing conditions become smaller. In such cases, the prediction model generation system 10 may decide to terminate the exploration at S600. The graph in Figure 10 is closer to the true data in Figure 5 than the previous graph.

[0120] According to the flow of this embodiment described above, the process of searching for manufacturing conditions, executing CAE based on the search results, and updating the learning model using the CAE results is repeated. This allows CAE to be executed preferentially for relatively important manufacturing conditions, thereby enabling the realization of a highly accurate predictive model while suppressing the consumption of computing resources. Furthermore, it reduces the need for humans to spend time deciding which manufacturing conditions should be subjected to CAE.

[0121] Figure 11 shows an example of the subflow of S300 in a modified example. In this modified example, the display unit 140 displays an unexplored planar region on a two-dimensional plane where the condition generation unit 132 has not generated manufacturing conditions. The display unit 140 may display the unexplored region on the two-dimensional plane in S335 after S330.

[0122] The unexplored area identification unit 138 identifies unexplored areas for which the condition generation unit 132 has not generated manufacturing conditions. The unexplored area identification unit 138 identifies separated points from any CAE-executed manufacturing conditions among the uniformly distributed points generated in S310 and / or the dummy data generated in S320, and supplies them to the display unit 140. The display unit 140 may identify the area formed by the separated points and display the area together with the CAE-executed manufacturing conditions, etc.

[0123] Figure 12 shows an example of displaying an unexplored region. For example, in S335, the display unit 140 may display an unexplored planar region, such as region 1210, along with the CAE-executed manufacturing conditions (e.g., display of X).

[0124] Here, the predictive model generation system 10 may receive instructions from the user to execute CAE under manufacturing conditions within an unexplored region. For example, if the user specifies a point within the displayed region 1210 by clicking or other means, the system receives the manufacturing conditions corresponding to that point as specified manufacturing conditions.

[0125] The CAE execution unit 110 immediately or in the next S500 simulates the manufacturing of a resin product under specified manufacturing conditions included within an unexplored plane region specified by the user, and obtains feature values. In the following S250, the model learning unit 120 updates the prediction model using the specified manufacturing conditions and the pair of feature values ​​of the resin product simulated by CAE under those specified manufacturing conditions.

[0126] Instead of the user specifying manufacturing conditions within the unexplored planar region, in the next S400, the condition generation unit 132 may generate manufacturing conditions included in the unexplored region as candidate manufacturing conditions or recommended manufacturing conditions. For example, the condition generation unit 132 may uniformly or randomly select a predetermined number of coordinates from the unexplored region identified by the unexplored identification unit 138 and use them as candidate manufacturing conditions or recommended manufacturing conditions. That is, the manufacturing conditions specified within the unexplored planar region may be used directly in the CAE of the CAE execution unit 110, or they may be used in the CAE after confirming the evaluation by the evaluation unit 136.

[0127] According to this modification example, it is possible to prevent the search for manufacturing conditions from falling into local optimization. According to this modification example, while prioritizing regions with high evaluation in the manufacturing condition space, CAE can also be executed for other regions to a certain extent. Therefore, a high-precision prediction model can be realized with few resources under a wide range of manufacturing conditions.

[0128] The prediction model generation system 10 only executes CAE for the manufacturing conditions of the unexplored region only at the initial stage of the search (for example, during the initial predetermined number of loop processes), and it is not necessary to identify the unexplored region after the initial stage of the search. By paying attention only to the unexplored region at the initial stage, the number of CAE executions can be further suppressed, and the risk of falling into local optimization can be reduced.

[0129] The prediction model generation system 10 only executes CAE for the manufacturing conditions of the unexplored region only at the later stage of the search (for example, after the predetermined number of loop processes), and it is not necessary to identify the unexplored region after the initial stage of the search. By paying attention to the unexplored region after a certain degree of search has progressed and after a bias has actually occurred in the distribution of the manufacturing conditions for which CAE has been executed, it is possible to avoid unnecessarily allocating resources to prevent local optimization.

[0130] Execute CAE for the manufacturing conditions of the unexplored region at the initial and later stages of the search, and it is not necessary to identify the unexplored region during the middle stage of the search. For example, it is possible to identify the unexplored region only during the loop processes after the nth time and up to the mth time (n and m are integers where n < m).

[0131] In the description of the above embodiment, an example that does not consider the anisotropy of the material contained in the resin product was described. In the second and third modification examples described below, the resin product includes an anisotropic material having anisotropy, and CAE includes simulating the orientation of the anisotropic material. In this modification, the characteristic value includes variables affected by the orientation of the anisotropic material. For example, the characteristic value includes at least one of the stress distribution, coefficient of thermal expansion distribution, and deformation of the resin product.

[0132] Anisotropic materials are materials that exhibit anisotropy in shape and / or physical properties. For example, anisotropic materials may be oriented materials, such as fibrous materials. One example of a fibrous material is glass fiber.

[0133] Figure 14 shows an example of anisotropy exhibited by anisotropic materials. Figure 14 shows the results of a simulation of the strain-stress curves for different orientation angles of glass fibers after injection molding of a resin material containing glass fibers. The horizontal axis represents strain (%), and the vertical axis represents stress (MPa). As shown, different curves are obtained for each orientation angle of the glass fibers (0 degrees, 20 degrees, 45 degrees, 90 degrees).

[0134] Figure 15 shows an example of stress acting on a small cubic region. In a small cubic region of a three-dimensional object defined by xyz coordinates, T xx ,T yy and T zz The three components of normal stress shown, and T xy ,T xz and T yz The three components of shear stress shown are acting. The prediction model generation system 10 of this modified example predicts the normal stress and shear stress as at least some of the feature values, taking into account the orientation of the anisotropic material.

[0135] Figure 16 shows the relationship between the prediction model and CAE in a second modified example of this embodiment. Based on the boundary conditions input in S100 as part of the fixed conditions (for example, load conditions and constraint conditions for the resin product), the CAE execution unit 110 performs CAE (structural analysis) 212, and characteristic values ​​in the non-oriented state (also called "characteristic values ​​(non-oriented state) 232") are output.

[0136] Feature value (unoriented state) 232 is a characteristic value such as stress distribution in the unoriented state of a resin material containing an oriented anisotropic material. The unoriented state is a state in which the anisotropic material does not exhibit anisotropy, and may be, for example, a state in which the anisotropic material takes on a random orientation. Feature value (unoriented state) 232 may include a tensor of feature values ​​in the unoriented state (e.g., a stress tensor).

[0137] For example, the CAE execution unit 110 performs CAE (structural analysis) 212 using an existing finite element analysis program such as Abaqus and obtains feature values ​​(unoriented state) 232.

[0138] The CAE execution unit 110 obtains orientation information 234 by performing CAE (flow analysis) 214 based on the manufacturing conditions. The orientation information 234 is information regarding the orientation of anisotropic materials contained in the resin material. For example, the orientation information 234 includes the orientation direction of anisotropic materials identified by the flow analysis.

[0139] For example, the CAE execution unit 110 may perform CAE (flow analysis) 214 using an existing injection molding simulation program such as Moldex3D to reproduce the flow of the resin material and obtain orientation information 234. As an example, the orientation information 234 may include an orientation tensor.

[0140] Manufacturing conditions input into CAE (Flow Analysis) 214 may include, for example, the shape of the resin product, the resin material (e.g., viscosity and / or specific heat), the structure of the manufacturing equipment (e.g., mold shape and / or gate position), and processing conditions (e.g., mold temperature, resin temperature, processing temperature and / or processing pressure).

[0141] Furthermore, the CAE execution unit 110 executes CAE (structural analysis) 216 based on the acquired feature values ​​(unoriented state) 232 and orientation information 234, thereby outputting feature values ​​of the resin product in an oriented state (also called "feature values ​​(orientation state) 236"). The CAE execution unit 110 may also output the orientation state 236 by executing CAE using the boundary conditions and orientation information 234 described above. As explained in Figures 14 and 15, stress distributions and the like can take on different values ​​depending on the orientation state. The feature values ​​(orientation state) 236 are feature values ​​recalculated taking into account the orientation information 234. For example, the CAE execution unit 110 generates the feature values ​​(orientation state) 236 by mapping the characteristics for each orientation, as shown in Figure 14, to the feature values ​​(unoriented state) 232 using an existing material modeling platform such as Digmat.

[0142] The CAE execution unit 110 may execute the above CAE at any timing in this embodiment and its modifications. For example, the CAE execution unit 110 may execute the above CAE (structural analysis) 212, CAE (fluid analysis) 214, and CAE (structural analysis) 216 at S200 and S500, etc., in the flow chart of Figure 3. The feature value (orientation state) 236 may include a tensor of feature values ​​in the orientation state (for example, a stress tensor).

[0143] Subsequently, the model learning unit 120 learns a predictive model based on the results of CAE (structural analysis) 212, CAE (flow analysis) 214, and CAE (structural analysis) 216. The model learning unit 120 learns a predictive model that takes manufacturing conditions as input and predicts the characteristic values ​​(orientation state) 236 of the resin product generated on CAE considering the orientation of anisotropic material based on those manufacturing conditions. At least some of the explanatory variables of the predictive model are the manufacturing conditions used in CAE (flow analysis) 214, and the objective variable is the characteristic values ​​(orientation state) 236 corresponding to those manufacturing conditions. For example, the model learning unit 120 may learn the predictive model at S250 in the flow chart of Figure 3.

[0144] As described above, according to the second modified example, the prediction unit 134 can predict the characteristic values ​​(orientation state) 236 of a resin product in an oriented state, generated on CAE, based on the characteristic values ​​(orientation state) 232 of a resin material containing an orientable anisotropic material in an orientable state, and orientation information 234 regarding the orientation of the anisotropic material contained in the resin material, using a prediction model. This allows the prediction model generation system 10 to perform searches while considering the anisotropy of the resin material.

[0145] Conventionally, it was often difficult to achieve both the molding conditions and the performance of the resin product itself. In this modified version, by considering the anisotropy of the resin material, it has become easier to achieve both the molding conditions and the performance of the product itself. According to this modified version, by considering the anisotropy of the resin material, the resin product can maintain sufficient strength with less material, contributing to weight reduction. According to this modified version, it has become possible to identify the weak points of the resin product by considering its orientation.

[0146] The second modified example describes an example in which only boundary conditions are input into CAE (structural analysis) 212, but it is not limited to this. In addition to boundary conditions, at least a portion of the manufacturing conditions may also be input into CAE (structural analysis) 212 and incorporated into the explanatory variables of the prediction model. Such manufacturing conditions may include the shape of the resin product and / or the resin material.

[0147] Boundary conditions may be incorporated as part of the manufacturing conditions. For example, at least a portion of the boundary conditions may be included as explanatory variables in the predictive model as part of the manufacturing conditions. This allows the boundary conditions to also be included in the exploration.

[0148] Figure 17 shows the relationship between the prediction model and CAE in the third modified example of this embodiment. This example differs from the second modified example in Figure 16 in that, instead of manufacturing conditions, feature values ​​(unoriented state) 232 and orientation information 234 are used as explanatory variables in the prediction model. That is, in this example, the prediction model takes as input the feature values ​​(unoriented state) 232 obtained by performing CAE (structural analysis) 212 using boundary conditions, and the orientation information 234 obtained by performing CAE (fluid analysis) 214 using manufacturing conditions, and predicts the feature values ​​(oriented state) 236. Therefore, in S250, the model learning unit 120 learns the prediction model using the set of feature values ​​(unoriented state) 232, orientation information 234, and feature values ​​(oriented state) 236 as learning data.

[0149] The prediction model may predict the feature value (orientation state) 236 from orientation information 234 only, without inputting the feature value (unoriented state) 232. In this case, the prediction model cannot adapt to changes in boundary conditions, so the model learning unit 120 may learn the prediction model for each boundary condition.

[0150] The prediction unit 134 uses boundary conditions to cause the CAE execution unit 110 to perform CAE (structural analysis) 212 and obtains characteristic values ​​(unoriented state) 232. The prediction unit 134 uses the input manufacturing conditions to cause the CAE execution unit 110 to perform CAE (flow analysis) 214 and obtains orientation information 234.

[0151] The prediction unit 134 obtains the feature value (orientation state) 236 by inputting the obtained feature value (unoriented state) 232 and orientation information 234 into the prediction model. In this example, since the orientation information 234 is obtained by CAE, the number of CAE (structural analysis) 216 executions can be reduced at least, while obtaining the feature value (orientation state) 236 more accurately compared to the second modified example.

[0152] Various embodiments of the present invention may be described with reference to flowcharts and block diagrams, where a block may represent (1) a stage in a process in which an operation is performed or (2) a section of a device having the role of performing the operation. Specific stages and sections may be implemented by dedicated circuits, programmable circuits supplied with computer-readable instructions stored on a computer-readable medium, and / or processors supplied with computer-readable instructions stored on a computer-readable medium. Dedicated circuits may include digital and / or analog hardware circuits, and may include integrated circuits (ICs) and / or discrete circuits. Programmable circuits may include reconfigurable hardware circuits, including logical AND, logical OR, logical XOR, logical NAND, logical NOR, and other logic operations, flip-flops, registers, memory elements such as field-programmable gate arrays (FPGAs), programmable logic arrays (PLAs), etc.

[0153] Computer-readable media may include any tangible device capable of storing instructions to be executed by a suitable device, and as a result, computer-readable media having instructions stored therein will comprise a product containing instructions that can be executed to create means for performing operations specified in a flowchart or block diagram. Examples of computer-readable media may include electronic storage media, magnetic storage media, optical storage media, electromagnetic storage media, semiconductor storage media, etc. More specific examples of computer-readable media may include floppy disks, diskettes, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), electrically erasable programmable read-only memory (EEPROM), static random access memory (SRAM), compact disk read-only memory (CD-ROM), digital multipurpose disc (DVD), Blu-ray® disc, memory stick, integrated circuit card, etc.

[0154] Computer-readable instructions may include assembler instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, state setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk®, Java®, C++, and traditional procedural programming languages ​​such as the C programming language or similar programming languages.

[0155] Computer-readable instructions may be provided locally or via a wide area network (WAN) such as a local area network (LAN) or the internet to the processor or programmable circuit of a programmable data processing device such as a general-purpose computer, a special-purpose computer, or another computer, and the computer-readable instructions may be executed to create means for performing operations specified in a flowchart or block diagram. Examples of processors include computer processors, processing units, microprocessors, digital signal processors, controllers, microcontrollers, etc.

[0156] Figure 13 shows an example of a computer 2200 in which multiple aspects of the present invention may be embodied in whole or in part. A program installed on the computer 2200 can cause the computer 2200 to function as an operation or one or more sections of an apparatus according to an embodiment of the present invention, or to execute such operation or one or more sections, and / or to cause the computer 2200 to execute a process or a stage of such process according to an embodiment of the present invention. Such a program may be executed by the CPU 2212 to cause the computer 2200 to perform a particular operation associated with some or all of the blocks in the flowcharts and block diagrams described herein.

[0157] The computer 2200 according to this embodiment includes a CPU 2212, RAM 2214, a graphics controller 2216, and a display device 2218, which are interconnected by a host controller 2210. The computer 2200 also includes input / output units such as a communication interface 2222, a hard disk drive 2224, a DVD-ROM drive 2226, and an IC card drive, which are connected to the host controller 2210 via an input / output controller 2220. The computer also includes legacy input / output units such as a ROM 2230 and a keyboard 2242, which are connected to the input / output controller 2220 via an input / output chip 2240.

[0158] The CPU 2212 operates according to programs stored in the ROM 2230 and RAM 2214, thereby controlling each unit. The graphics controller 2216 retrieves image data generated by the CPU 2212 from a frame buffer provided in RAM 2214 or from itself, and displays the image data on the display device 2218.

[0159] The communication interface 2222 communicates with other electronic devices via a network. The hard disk drive 2224 stores programs and data used by the CPU 2212 in the computer 2200. The DVD-ROM drive 2226 reads programs or data from the DVD-ROM 2201 and provides them to the hard disk drive 2224 via the RAM 2214. The IC card drive reads programs and data from the IC card and / or writes programs and data to the IC card.

[0160] The ROM 2230 stores boot programs and / or programs that depend on the computer 2200's hardware, which are executed by the computer 2200 when activated. The input / output chip 2240 may also connect various input / output units to the input / output controller 2220 via parallel ports, serial ports, keyboard ports, mouse ports, etc.

[0161] The program is provided on a computer-readable medium such as a DVD-ROM 2201 or an IC card. The program is read from the computer-readable medium and installed on a hard disk drive 2224, RAM 2214, or ROM 2230, which are also examples of computer-readable medium, and executed by the CPU 2212. The information processing described within these programs is read by the computer 2200, resulting in coordination between the program and the various types of hardware resources described above. The apparatus or method may be configured to realize the manipulation or processing of information in accordance with the use of the computer 2200.

[0162] For example, when communication is performed between a computer 2200 and an external device, the CPU 2212 may execute a communication program loaded into RAM 2214 and, based on the processing described in the communication program, instruct the communication interface 2222 to perform communication processing. Under the control of the CPU 2212, the communication interface 2222 reads transmission data stored in a transmission buffer processing area provided in a recording medium such as RAM 2214, a hard disk drive 2224, a DVD-ROM 2201, or an IC card, transmits the read transmission data to the network, or writes received data received from the network to a reception buffer processing area provided on the recording medium.

[0163] The CPU 2212 reads all or necessary parts of a file or database stored on an external storage medium such as a hard disk drive 2224, a DVD-ROM drive 2226 (DVD-ROM 2201), or an IC card into the RAM 2214, and may perform various types of processing on the data in the RAM 2214. The CPU 2212 then writes the processed data back to the external storage medium.

[0164] Various types of information, such as various types of programs, data, tables, and databases, may be stored on the recording medium and subjected to information processing. The CPU 2212 may perform various types of processing on the data read from RAM 2214, including various types of operations, information processing, conditional judgments, conditional branching, unconditional branching, information retrieval / replacement, etc., as described throughout this disclosure and specified by the program instruction sequence, and write the results back to RAM 2214. The CPU 2212 may also retrieve information in files, databases, etc., within the recording medium. For example, if multiple entries are stored in the recording medium, each having an attribute value of a first attribute associated with an attribute value of a second attribute, the CPU 2212 may search among the multiple entries for an entry that matches the condition for which the attribute value of the first attribute is specified, read the attribute value of the second attribute stored in that entry, and thereby obtain the attribute value of the second attribute associated with the first attribute that satisfies a predetermined condition.

[0165] The programs or software modules described above may be stored on or near computer 2200 on a computer-readable medium. Alternatively, recording media such as hard disks or RAM provided within a server system connected to a dedicated communication network or the Internet can be used as computer-readable media, thereby providing programs to computer 2200 via the network.

[0166] Although the present invention has been described above using embodiments, the technical scope of the present invention is not limited to the scope described in the above embodiments. It will be apparent to those skilled in the art that various modifications or improvements can be made to the above embodiments. It will be clear from the claims that such modified or improved forms may also be included in the technical scope of the present invention.

[0167] It should be noted that the execution order of operations, procedures, steps, and stages in the devices, systems, programs, and methods shown in the claims, specifications, and drawings is not explicitly stated as "before," "prior to," etc., and can be implemented in any order unless the output of a previous process is used in a later process. Even if the operation flow in the claims, specifications, and drawings is described using phrases such as "first," "next," etc. for convenience, it does not mean that it is essential to perform them in that order. The notation "A and / or B" may mean "A, B, or A and C." The notation "A, B and / or C" may mean "any one of A, B, and C, or any combination of two or more of these." [Explanation of Symbols]

[0168] 10. Predictive Model Generation System 110 CAE Execution Unit 120 Model Learning Department 130 Search Department 132 Condition generator 134 Prediction Section 136 Evaluation Department 138 Unexplored specific part 140 Display section 210 CAE 212 CAE (Structural Analysis) 214 CAE (Flow analysis) 216 CAE (Structural Analysis) 220 Predictive Models 222 Predictive Models 224 Predictive Models 232 Feature values ​​(unoriented state) 234 Orientation Information 236 Feature Values ​​(Orientation State) 610 areas 710 area 810 area 820 areas 910 area 1010 area 1210 area 2200 Computers 2201 DVD-ROM 2210 Host Controller 2212 CPU 2214 RAM 2216 Graphics Controller 2218 Display Devices 2220 Input / Output Controller 2222 Communication Interface 2224 Hard Disk Drive 2226 DVD-ROM drive 2230 ROM 2240 Input / Output Chip 2242 keyboard

Claims

1. A search unit searches for manufacturing conditions that satisfy predetermined conditions by using a predictive model that predicts the characteristic values ​​of a resin product generated on CAE (Computer-Aided Engineering) using the manufacturing conditions of the resin product, based on the manufacturing conditions of the resin product. A model learning unit updates the prediction model using data including the manufacturing conditions obtained through exploration and the characteristic values ​​of the resin product simulated by CAE using those manufacturing conditions. A predictive model generation system equipped with the following features.

2. The search unit, The aforementioned prediction model has a prediction unit that inputs manufacturing conditions and outputs characteristic values. The predictive model generation system according to claim 1.

3. The search unit, It has a condition generation unit that generates manufacturing conditions used for exploration, The predictive model generation system according to claim 2.

4. The search unit, The predictive model has an evaluation unit that inputs the manufacturing conditions and evaluates the obtained characteristic values. The predictive model generation system according to claim 3.

5. The evaluation unit described above, Output evaluation values ​​obtained by evaluating feature values. The prediction unit searches for manufacturing conditions under which it predicts an evaluation value above a threshold. The predictive model generation system according to claim 4.

6. The evaluation unit searches for and outputs multiple manufacturing conditions that give an evaluation value equal to or greater than a threshold. The predictive model generation system according to claim 5.

7. The condition generation unit applies a genetic algorithm to multiple manufacturing conditions that give an evaluation value above a threshold to generate new manufacturing conditions. The predictive model generation system according to claim 6.

8. The evaluation unit searches for and outputs multiple manufacturing conditions that give a Pareto solution in the feature space defined by the numerical values ​​of the feature values. The predictive model generation system according to claim 5.

9. The search unit, In the manufacturing condition space defined by the numerical values ​​of the manufacturing conditions, the condition generation unit has an unexplored identification unit that identifies unexplored regions where the condition generation unit has not generated manufacturing conditions. The condition generation unit generates manufacturing conditions included in the unexplored region. The predictive model generation system according to claim 3.

10. The evaluation unit uses the prediction model updated by the model learning unit to perform an additional search to further search for manufacturing conditions under which the prediction unit predicts an evaluation value equal to or greater than a threshold. The model learning unit performs an additional update to update the prediction model using the manufacturing conditions obtained through the additional search and the pair of feature values ​​of the resin product simulated by CAE using those manufacturing conditions. The predictive model generation system according to claim 5.

11. The above additional search and the above additional update processes are repeated, The evaluation unit described above, In the aforementioned iteration, the iteration is terminated when the movement of the manufacturing conditions in the manufacturing condition space defined by the numerical values ​​of the manufacturing conditions, for which the prediction unit predicts an evaluation value above a threshold, falls below the threshold. The predictive model generation system according to claim 10.

12. The manufacturing conditions include at least one piece of information regarding the material of the resin product and the processing conditions of the resin product. The predictive model generation system according to claim 1.

13. The aforementioned characteristic value includes at least one of the following: performance and / or physical properties of the resin product, numerical values ​​relating to its shape, and numerical values ​​relating to the conditions necessary for its manufacture. The predictive model generation system according to claim 1.

14. The aforementioned resin product is an injection-molded resin product. The CAE includes simulating the injection molding process of the resin product, The predictive model generation system according to claim 1.

15. The aforementioned resin product includes an anisotropic material having anisotropy, The CAE includes simulating the orientation of the anisotropic material, The predictive model generation system according to claim 1.

16. The prediction model takes orientation information regarding the orientation of the anisotropic material obtained by performing the CAE using the manufacturing conditions as input and predicts the feature values. The predictive model generation system according to claim 15.

17. The prediction model takes the manufacturing conditions as input and predicts the characteristic values ​​of the resin product generated on the CAE considering the orientation of the anisotropic material based on the manufacturing conditions. The predictive model generation system according to claim 15.

18. The aforementioned anisotropic material is a fibrous material. The predictive model generation system according to claim 15.

19. The system includes a display unit that compresses the manufacturing condition space, defined by numerical values ​​of the manufacturing conditions, into a two-dimensional plane, and displays the manufacturing conditions searched by the search unit on the two-dimensional plane. The predictive model generation system according to claim 3.

20. The display unit displays an unexplored planar region on the two-dimensional plane in which the condition generation unit has not generated manufacturing conditions. The predictive model generation system according to claim 19.

21. The aforementioned model learning unit, The predictive model is updated using the specified manufacturing conditions included within the unexplored plane region specified by the user, and the pair of feature values ​​of the resin product simulated by CAE using the specified manufacturing conditions. The predictive model generation system according to claim 20.

22. The system further includes a CAE execution unit that simulates the manufacturing of resin products using the manufacturing conditions obtained through exploration, The predictive model generation system according to claim 1.

23. It is executed by a computer, and the computer, To function as a predictive model generation system according to any one of claims 1 to 22, program.

24. Based on the manufacturing conditions of the resin product, a prediction model is used to predict the characteristic values ​​of the resin product generated on CAE (Computer-Aided Engineering) using those manufacturing conditions, and a search stage is performed to find manufacturing conditions that give characteristic values ​​that satisfy predetermined conditions. A model learning stage in which the prediction model is updated using the manufacturing conditions obtained through exploration and the pair of feature values ​​of the resin product simulated by CAE using said manufacturing conditions, A predictive model generation method comprising the following features.

25. The system includes a prediction unit that uses a prediction model to predict the characteristic values ​​of a resin product in a state where the anisotropic material is oriented, based on the characteristic values ​​of a resin material containing an orientable anisotropic material in an unoriented state, and orientation information regarding the orientation of the anisotropic material contained in the resin material, generated on CAE (Computer-Aided Engineering). Prediction system.

26. The aforementioned characteristic value is a variable that is affected by the orientation of the anisotropic material, The orientation information includes the orientation direction of the anisotropic material identified by flow analysis. The prediction system according to claim 25.

27. The aforementioned characteristic values ​​include at least one of the stress distribution, thermal expansion coefficient distribution, or deformation of the resin product. The prediction system according to claim 26.

28. It is executed by a computer, and the computer, To function as a prediction system according to any one of claims 25 to 27, program.

29. A prediction step in which a predictive model is used to predict the characteristic values ​​of a resin product in a state in which the anisotropic material is oriented, based on the characteristic values ​​of a resin material containing an orientable anisotropic material in an unoriented state, and orientation information regarding the orientation of the anisotropic material contained in the resin material, generated on CAE (Computer-Aided Engineering); A prediction method that includes the following features.

Images