Plate thickness optimization method and plate thickness optimization program

Abstract

This invention provides a plate thickness optimization method that enables efficient and highly accurate plate thickness design. [Solution] A method for optimizing the thickness of a resin member, comprising: a first shape deformation step of generating multiple first learning samples by deforming a model 10 of a resin member partitioned into multiple meshes 20 by modeling; a first normalization execution step of performing first normalization on the multiple first learning samples based on the displacement amount, maximum value and minimum value of the displacement amount for each mesh 20; a second shape deformation step of generating multiple second learning samples that fall within a predetermined thickness range based on the first normalization values ​​obtained in the first normalization execution step; a performance index analysis step of performing structural analysis using the finite element method on the second learning samples and analyzing performance indices corresponding to a predetermined deformation amount; and a learning data generation step of generating multiple second learning samples and performance indices as a learning dataset.

Description

[Technical Field]

[0001] The present invention relates to a plate thickness optimization method and a plate thickness optimization program for resin members used in automobiles and the like. [Background Art]

[0002] Conventionally, various resin members have been used in various automobiles and the like. When designing such a resin member, a designer performs design while conducting analysis using CAE (Computer Aided Engineering) (see, for example, Patent Document 1). [Prior Art Documents] [Patent Documents]

[0003] [Patent Document 1] Japanese Unexamined Patent Application Publication No. 2005-144881 [Summary of the Invention] [Problems to be Solved by the Invention]

[0004] However, in the conventional technology as described in the aforementioned Patent Document 1, although a designer performs design using CAE, in order to design a plate thickness that satisfies requirements such as moldability, weight reduction, and heat resistance, it has been necessary to perform repeated calculations based on model creation that relies on the designer's experience and intuition and CAE analysis results to narrow down to the optimal plate thickness. For this reason, there has been a problem that the man-hour burden associated with design increases.

[0005] Accordingly, an object of the present invention is to provide a plate thickness optimization method that enables efficient design of resin members and the like, and also enables highly accurate plate thickness design. [Means for Solving the Problems]

[0006] (1) A plate thickness optimization method provided to solve the above-mentioned problems is a plate thickness optimization method for searching for the optimal plate thickness of a resin member, and is characterized by comprising: a first shape deformation step of deforming a model of the resin member partitioned into a plurality of meshes by modeling to generate a plurality of first learning samples; a first normalization execution step of performing a first normalization on the plurality of first learning samples based on the displacement amount for each mesh, and the maximum and minimum values ​​of the displacement amount; a second shape deformation step of generating a plurality of second learning samples so that they fall within a predetermined plate thickness range based on the first normalization values ​​calculated in the first normalization execution step; a performance index analysis step of performing a structural analysis by finite element method on the plurality of second learning samples to analyze performance indexes corresponding to a predetermined deformation amount; and a learning data generation step of generating the plurality of second learning samples and the performance indexes as a learning dataset.

[0007] The plate thickness optimization method of the present invention described above involves performing a first normalization on the first learning sample based on the displacement amount for each mesh, as well as the maximum and minimum values ​​of the displacement amount, in the first normalization execution step. As a result, the plate thickness optimization method of the present invention can normalize the distribution range of plate thickness and scale it to a predetermined range, thereby reducing the range of variation and enabling the execution of the second shape deformation step. Furthermore, since comparisons can be made on the same scale for each sample, accurate analysis can be performed.

[0008] For example, when forming a resin component by injection molding, the resin is introduced through the gate. As a result, the degree of thermal deformation differs between the gate area (also called the gate position) and the general surface (excluding the gate position) of the molded resin component, and therefore the thickness distribution tends to differ as well. Consequently, when deforming the model, if morphing is performed across the entire model, there is a concern that the thickness near the gate will not be constant after normalization. To solve this, the inventors diligently studied the matter and came to the conclusion that in the first morphing, the amount of thickness displacement is obtained across the entire model and the first normalization (also called first normalization) is performed, and in the second morphing, the gate position is fixed within a predetermined range (for example, zero), and the second normalization (also called second normalization) is performed using scaling from the first normalization.

[0009] (2) The plate thickness optimization method of the present invention described above is characterized in that the resin member is formed by injecting resin from a gate by injection molding, and the second shape deformation step is characterized in that, when deforming the model, the amount of deformation at the position corresponding to the gate in the model is limited to a predetermined range, and a second normalization is performed based on the maximum and minimum values ​​of the displacement for each mesh obtained in the first shape deformation step so that it becomes a predetermined plate thickness range, thereby generating a plurality of second learning samples.

[0010] The plate thickness optimization method of the present invention, as described above, can be configured as in (2) above to keep the plate thickness deviation constant by considering the formability of the general surface (excluding the area near the gate) and the area near the gate. Therefore, the plate thickness optimization method of the present invention can optimize the plate thickness with high accuracy. In other words, the plate thickness optimization method of the present invention, as described in (2) above, can perform second normalization using the scaling of first normalization, so that displacement amounts such as plate thickness can be compared and evaluated within the same scaling range. As a result, the plate thickness optimization method of the present invention can generate highly accurate analysis data regarding displacement amounts as a second training sample. Here, the range of deformation at the position corresponding to the gate in the model (also referred to as the gate position) can be limited to various ranges, but for example, it is good to fix it to zero. As a result, the plate thickness optimization method of the present invention can obtain an optimal value of plate thickness that satisfies the target performance indicators (e.g., weight and heat resistance) from the gate to the general surface while keeping the plate thickness near the gate constant.

[0011] (3) The plate thickness optimization method of the present invention described above is characterized by having an AI program that explores the relationship between the plate thickness of the resin member and the performance index, and performing an AI learning step of training the AI ​​program with the training dataset, and an optimization search step of exploring the relationship between the plate thickness and the performance index based on the trained AI program.

[0012] The plate thickness optimization method of the present invention described above allows an AI program to be trained using a dataset of multiple second training samples and performance indicators, and based on the trained AI program, it can explore the relationship between plate thickness and the performance indicators. Therefore, the plate thickness optimization method of the present invention can perform efficient plate thickness optimization exploration using AI. As a result, the plate thickness optimization method of the present invention can efficiently perform stable and accurate plate thickness design without relying on the experience or intuition of the designer, and thus a reduction in design man-hours can be expected.

[0013] (4) The plate thickness optimization method of the present invention described above is preferably characterized in that at least one of the first shape deformation step and the second shape deformation step deforms the model based on design of experiments.

[0014] The plate thickness optimization method of the present invention described above performs model deformation based on experimental design, so it can efficiently generate first and second training samples without unnecessarily increasing the number of samples. As a result, the plate thickness optimization method of the present invention can obtain accurate analysis results while speeding up the CAE analysis.

[0015] (5) The plate thickness optimization method of the present invention described above is preferably characterized in that the performance index is based on one or more of the following: heat resistance, cold resistance, rigidity, fluidity, and impact resistance.

[0016] The plate thickness optimization method of the present invention described above, when configured as described in (5) above, can optimize plate thickness design based on performance indicators that affect plate thickness. Here, the performance indicators may include abrasion resistance, scratch resistance, paintability, thermal conductivity, and weather resistance.

[0017] Here, if the morphing deformation is not performed smoothly, there was a concern that the change in plate thickness for each mesh would be stepwise, resulting in large changes in plate thickness.

[0018] (6) The plate thickness optimization method of the present invention described above is preferably characterized in that the deformation of the model in the first shape deformation step and the second shape deformation step is performed by morphing deformation based on a spline function.

[0019] In the plate thickness optimization method of the present invention described above, the deformation of the model is performed by morphing deformation based on a spline function, so the change in the amount of plate thickness displacement for each mesh can be changed along the spline curve. As a result, the plate thickness optimization method of the present invention can smooth the change in plate thickness, and thus enable highly accurate optimization of plate thickness.

[0020] (7) In the plate thickness optimization method of the present invention described above, it is preferable that the algorithm in the AI program is executed based on a graph neural network.

[0021] The plate thickness optimization method of the present invention described above, configured as in (7) above, enables high-precision plate thickness optimization using a graph neural network (GNN).

[0022] (8) In the plate thickness optimization method of the present invention described above, it is preferable that the graph neural network uses the adjacency matrix of each node as an explanatory variable.

[0023] The plate thickness optimization method of the present invention described above, configured as in (8) above, can efficiently convert the relationship between a plurality of meshes into explanatory variables.

[0024] (9) In the plate thickness optimization method of the present invention described above, it is preferable that the graph neural network uses, as an explanatory variable for each node, a plate thickness value calculated by adding the plate thicknesses of the plurality of meshes corresponding to each directly connected node.

[0025] The plate thickness optimization method of the present invention described above, configured as in (9) above, can be efficiently obtained as an explanatory variable relating to plate thickness values between a plurality of meshes.

[0026] (10) The plate thickness optimization program of the present invention described above is preferably capable of executing the above (1) to (9) in an information processing apparatus.

[0027] The plate thickness optimization program of the present invention is executable in an information processing apparatus, so it can be executed in information processing apparatuses such as various computers. Therefore, according to the plate thickness optimization program of the present invention, versatility is improved. Effects of the Invention

[0028] According to the present invention, it is possible to provide a plate thickness optimization method that enables efficient design of resin components and the like, as well as accurate plate thickness design. [Brief explanation of the drawing]

[0029] [Figure 1] This is an explanatory diagram of a target model for plate thickness design using a plate thickness optimization method according to one embodiment of the present invention, and shows the process when the first morphing operation is performed. [Figure 2] This is an explanatory diagram for a target model used in plate thickness design by a plate thickness optimization method according to one embodiment of the present invention, specifically for the second morphing (fixing near the gate) operation. [Figure 3] (a) is an explanatory diagram showing the thickness distribution of the target model at the start of thickness design in the thickness optimization method of the present invention, and (b) is an explanatory diagram showing the thickness distribution of the target model after the thickness design is completed in the thickness optimization method of the present invention. [Figure 4] (a) and (b) are explanatory diagrams illustrating the morphing of the general surface (surfaces other than the gate portion) of the target model in the plate thickness optimization method of the present invention. [Figure 5] (a) and (b) are explanatory diagrams illustrating the morphing near the gate position of the target model in the plate thickness optimization method of the present invention. [Figure 6] (a) is an explanatory diagram representing some of the elements of the meshed model, (b) is an explanatory diagram representing the adjacency matrix of (a), and (c) is an explanatory diagram representing the features (graph: explanatory variables) of each node in (a). [Figure 7] This is a flowchart of a plate thickness optimization method according to one embodiment of the present invention. [Figure 8] This flowchart follows Figure 7. [Figure 9] This is a flowchart following Figure 8. [Modes for carrying out the invention]

[0030] The following describes in detail a plate thickness optimization method according to one embodiment of the present invention, with reference to the drawings. Note that the figures are schematic representations for ease of understanding and may differ from the actual shape, size, and arrangement of components. Also, similar reference numerals for components may be omitted in the figures.

[0031] As shown in Figures 1 to 3, this embodiment describes a case in which the plate thickness optimization method of the present invention is applied to the plate thickness design of a resin component (also referred to as a resin molded product) that constitutes the back door garnish of an automobile. Furthermore, this embodiment describes as an example in which the plate thickness optimization method is used to optimize the plate thickness of a resin component that achieves both weight reduction and heat resistance. First, in explaining the plate thickness optimization method of the present invention, a general flow will be described with reference to Figures 1 to 6. Note that in Figures 1 and 2, the mesh 20, node 21, morphing box 30, and morphing point 31 may be omitted except in some cases.

[0032] As shown in Figure 1, the modeling of the resin component is performed by an information processing device (not shown), such as a personal computer. In this modeling, the external shape of a resin component made from the same resin raw material is formed as model 10. Model 10 is divided (partitioned) into a mesh, and the intersections of each mesh 20 are set as nodes 21. Model 10 also has a gate 11, which is the starting point for resin flow during injection molding, and general surfaces 12 other than the gate 11.

[0033] Here, the information processing device is capable of performing CAE analysis based on a built-in finite element method-based structural analysis program (CAE). Furthermore, the information processing device is capable of performing optimization searches based on performance indicators such as thermal deformation temperature, thermal deformation amount, stiffness, fluid analysis, and impact resistance using a built-in AI program. The AI ​​program can be trained using the second training sample and training dataset generated based on the performance indicators described later. The CAE and AI programs may be configured to work together as needed, so that the search results from the AI ​​program are reflected in the CAE.

[0034] Once the model 10 is formed, a cubic morphing box 30 is set up to surround the mesh 20 to be deformed, and morphing points 31 (for example, vertices of the morphing box 30) corresponding to the morphing box 30 are set. In addition, morphing points 31 to be manipulated based on the design of experiments are set. In this embodiment, based on the design of experiments, data is generated that assigns, for example, plate thickness to thermal deformation temperature and data that assigns plate thickness to weight, and the amount of morphing (amount of deformation) for each morphing point 31 is determined based on this data.

[0035] Once the morphing amount for the morphing point 31 to be manipulated is set, a CAE analysis is performed as shown in Figure 3(a), and the first morphing (also referred to as the first shape deformation step) is executed based on the set morphing point 31. As a result, the morphing box 30 is deformed, and each mesh 20 deforms in accordance with the deformation of the morphing box 30. In addition, multiple first learning samples corresponding to each mesh 20 are generated along with the deformation of each mesh 20. Here, the first shape deformation step is assumed to be a morphing deformation based on a spline function. As a result, the morphing is executed smoothly based on the spline function, so that the displacement amount of the plate thickness is interpolated smoothly rather than in a step-like manner. Note that the hatched area in Figure 3 represents the distribution of the plate thickness displacement amount, and in reality represents the distribution of gradational bumps and dips by color, but for convenience it is shown as solid hatching.

[0036] Figure 4 shows the distribution of plate thickness displacement amounts for multiple meshes 20 corresponding to the general surface 12 in Model 10. Figure 5 shows the distribution of plate thickness displacement amounts (corresponding to the first learning sample) for multiple elements (mesh 20) corresponding to the position corresponding to the gate 11 in Model 10 (also referred to as gate position 11). As shown in Figures 4(a) and 5(a), when the first shape deformation step is performed, the CAE analysis generates distributions corresponding to the plate thickness displacement amounts at the general surface 12 and gate 11. Once the distribution of plate thickness displacement amounts is generated, the scale for normalization (first normalization) is determined based on the maximum and minimum values ​​of the plate thickness displacement amounts, and normalization (also referred to as the first normalization execution step) is performed. This generates first normalized values.

[0037] Next, as shown in Figure 2, the amount of morphing at the gate position 11 enclosed by the dashed-dot line is limited to a predetermined range (zero in this embodiment), and the amount of morphing at the other general surfaces 12 is determined based on the experimental design method, as described above. Based on the determined amount of morphing, morphing points 31 corresponding to each morphing box 30 are set. Once the morphing points 31 are set, a second morphing (also referred to as the second shape deformation step) is performed. Here, in the second shape deformation step, morphing deformation based on a spline function is performed. As a result, the morphing is performed smoothly based on the spline function, and the amount of plate thickness displacement is interpolated smoothly rather than in a step-like manner.

[0038] Furthermore, in the second shape deformation step, as shown in Figures 4(b) and 5(b), when deforming the model 10, the amount of deformation at the position corresponding to the gate 11 in the model 10 is limited to a predetermined range (for example, zero displacement). Also, based on the maximum and minimum displacement values ​​for each mesh 20 obtained in the first shape deformation step, a second normalization (also referred to as the second normalization execution step) is performed so that it falls within a predetermined plate thickness range. As a result, multiple second learning samples corresponding to each mesh 20 are generated so that they fall within a predetermined plate thickness range based on the first normalization value calculated in the first normalization execution step. In this embodiment, when generating the second learning samples, 2 (mm) is added to the total displacement amount for each mesh 20 generated in the second normalization execution step. As a result, second learning samples are generated in which the range of the normalized scale is shifted by 2 (mm). The reason for adding 2 (mm) to the plate thickness displacement is to set the plate thickness displacement to the range of 2 to 3 mm, which is the target value for the average plate thickness of the resin member (model 10) in this embodiment. Specifically, in the normalized scale, the minimum value is 0 and the maximum value is 1. Therefore, in this embodiment, to make the minimum value of the plate thickness displacement 2 and the maximum value 3, 2 is added to the result of the normalization process. For example, if you want a minimum value of 1.5 and a maximum value of 2, you should add 1.5 to the plate thickness multiplied by 0.5 after the normalization process. Also, for example, if you want a minimum value of 1 and a maximum value of 3, you should add 1 to the plate thickness multiplied by 2 after the normalization process.

[0039] Next, structural analysis using the finite element method is performed on multiple second training samples to analyze performance indices corresponding to a predetermined amount of deformation (e.g., thermal deformation temperature, relationship between weight and plate thickness) (also referred to as the performance index analysis step). The performance indices generated in the performance index analysis step are combined with the multiple second training samples to create a training dataset (also referred to as the training data generation step).

[0040] Here, we will explain a specific example of generating a training dataset, referring to Figure 6. Please note that Figure 6 only shows a portion of mesh 20.

[0041] As shown in Figure 6(a), when generating a training dataset, first, nodes 21, which are the intersections of each mesh 20, are extracted. In the figure, a rectangular mesh 20A and a triangular mesh 20B are shown adjacent to each other (other meshes 20 are not shown). Nodes 21 are sequentially assigned management numbers (e.g., 1 to n). In this embodiment, nodes 21, 21 corresponding to management numbers 3 and 4 are shared by mesh 20A and mesh 20B.

[0042] In the information processing device, once each node 21 is extracted, as shown in Figure 6(b), each node 21 is matrix-formed, generating an adjacency matrix that constitutes part of the explanatory variables. In the adjacency matrix, adjacent nodes 21 are assigned a flag 1, and nodes 21 that are not adjacent to each other are assigned a flag 0. This establishes the relationships between each mesh 20 (mesh 20A and mesh 20B in the figure).

[0043] Next, as shown in Figure 6(c), features (in this embodiment, coordinates, constraint presence / absence, and plate thickness) that constitute part of the explanatory variables are assigned to each node 21. Here, the presence / absence of constraints indicates, for example, whether or not morphing (deformation) at the gate position 11 is allowed; a flag 1 is assigned if it is allowed, and a flag 0 is assigned if it is not allowed. The plate thickness indicates, for example, the plate thickness of the mesh 20 corresponding to each node 21. In this embodiment, if there are multiple meshes 20 corresponding to each node 21, the plate thickness of all corresponding meshes 20 is added together. Therefore, for example, in nodes 21, 21 corresponding to management numbers 3 and 4, the plate thickness is input (calculated) as the sum of the plate thicknesses of mesh 20A (plate thickness: 2 mm) and mesh 20B (plate thickness: 1 mm) (3 mm).

[0044] As described above, once explanatory variables are input for each node 21, a second training sample is generated based on these explanatory variables. Once the second training sample is generated, the information processing device performs structural analysis using the finite element method (CAE analysis), and a performance index corresponding to a predetermined amount of deformation is analyzed (also referred to as the performance index analysis step).

[0045] The performance indicators obtained from the analysis are input to the AI ​​program in the information processing device as a training dataset along with the second training sample, and are used to train the AI ​​program (also referred to as the AI ​​training step). In this embodiment, since the training dataset is generated by explanatory variables called graphs, the AI ​​program is executed by an algorithm also called a graph neural network (GNN). Based on the training results, the AI ​​program performs a search to optimize the relationship between plate thickness and performance indicators (also referred to as the optimization search step). Furthermore, the AI ​​program can further optimize the relationship between plate thickness and performance indicators by learning from multiple training datasets based on the results of multiple structural analyses.

[0046] Next, with reference to Figures 7 to 9, a flowchart of one embodiment of the plate thickness optimization method of the present invention will be described.

[0047] As shown in Figure 7, first, meshing (mesh creation) is performed on the model 10 (step S1). Next, a morphing box 30 is set up for the model 10 to perform morphing as the first shape deformation step (step S2). Here, a morphing box 30 is set up for each of the multiple morphing points 31. Next, the number of morphing points (also called the number of elements) is obtained (step S3).

[0048] Next, based on the experimental design method, the morphing displacement amount of the morphing point 31 in each sample is determined, and a morphing matrix (first morphing matrix) is generated (step S10). Also, the base file corresponding to the old morphing matrix for the sample before morphing is opened (step S11), and the first morphing (first shape deformation step) is performed based on the first morphing matrix (step S12). The results are saved in the base file. That is, the execution result of the old morphing matrix in the base file is updated with the execution result of the first morphing matrix. Next, the displacement amount of each morphing point (element) is obtained from the updated base file (step S13). Steps S11 to S13 are repeated according to the number of samples.

[0049] Next, the maximum and minimum plate thickness displacement values ​​for each element are calculated (step S14). Then, the plate thickness displacement near gate position 11 in the first morphing matrix is ​​limited (fixed) to zero, thereby generating the second morphing matrix (step S20).

[0050] Next, as shown in Figure 8, a base file relating to the execution result of the first morphing matrix corresponding to the sample is opened (step S21), and a second morphing (second shape deformation step) is performed based on the second morphing matrix (step S22). The results are saved in the base file. That is, the execution result of the first morphing matrix in the base file is updated with the execution result of the second morphing matrix. Subsequently, the displacement amount of each morphing point (element) is obtained from the updated base file (step S23). Steps S21 to S23 are repeated according to the number of samples.

[0051] Next, the plate thickness displacement is first normalized using the displacement matrix of each element, and scaling is performed on the plate thickness matrix (step S30, first normalization execution step). Subsequently, the base file related to the execution result of the old plate thickness matrix is ​​opened (step S31), and the plate thickness is updated based on the plate thickness matrix, and the base file related to the execution result of the old plate thickness matrix is ​​updated (step S32). Steps S31 to S33 are repeated depending on the number of samples.

[0052] In step S32, when the plate thickness is updated, a structural analysis using CAE is performed (step S33). Furthermore, the thermal deformation temperature at which uneven deformation occurs is obtained from the CAE output data (step S34). Steps S31 to S34 are repeated depending on the number of samples.

[0053] Next, the input data from the CAE (FE mesh: finite element method mesh) is converted into a graph (explanatory variables) (step S40). Step S40 is repeated according to the number of samples (number of nodes).

[0054] Next, as shown in Figure 9, the explanatory variables (graphs) and the target variable (thermal distortion temperature) are split into training and validation sets (step S41). The training data, which has been split, is then input into the AI ​​program. As a result, the AI ​​program learns based on the training data (step S42). Subsequently, the accuracy is validated using the validation data (step S43). Steps S41 to S43 are repeated depending on the number of samples (training dataset).

[0055] Next, multi-objective optimization (e.g., plate thickness optimization) is performed to maximize the thermal distortion temperature and minimize the weight of the product (resin component) (step S44). In conjunction with this, a plate thickness distribution that optimizes the thermal distortion temperature and product weight is obtained (generated) (step S45). Once step S45 is completed, the process ends. Steps S1 to S45 may be repeated as needed.

[0056] The above is one embodiment of the plate thickness optimization method of the present invention. Next, the effects and advantages realized by the plate thickness optimization method of the present invention will be described in detail below.

[0057] The plate thickness optimization method of the present invention described above has the following characteristic configurations (a) to (j). Therefore, the plate thickness optimization method of the present invention can achieve unique effects that cannot be achieved by the prior art, as described below.

[0058] (a) The plate thickness optimization method of the embodiment described above is a plate thickness optimization method for searching for the optimal plate thickness of a resin member, and is characterized by comprising: a first shape deformation step of generating a plurality of first learning samples by deforming a model 10 of the resin member partitioned into a plurality of meshes 20 by modeling; a first normalization execution step of performing a first normalization on the plurality of first learning samples based on the displacement amount for each mesh 20, and the maximum and minimum values ​​of the displacement amount; a second shape deformation step of generating a plurality of second learning samples so that they fall within a predetermined plate thickness range based on the first normalization values ​​calculated in the first normalization execution step; a performance index analysis step of analyzing performance indexes corresponding to a predetermined deformation amount by performing a structural analysis using the finite element method on the plurality of second learning samples; and a learning data generation step of generating the plurality of second learning samples and the performance index as a learning dataset.

[0059] The plate thickness optimization method of the embodiment described above performs a first normalization execution step on the first learning sample based on the displacement amount for each mesh 20, as well as the maximum and minimum values ​​of the displacement amount. As a result, the plate thickness optimization method of the embodiment can normalize the distribution range of plate thickness and scale it to a predetermined range, so that the range of variation can be reduced while executing the second shape deformation step. Furthermore, since comparisons can be made on the same scale for each sample, accurate analysis can be performed.

[0060] Here, for example, when forming a resin component by injection molding, the resin is introduced from the gate 11. As a result, the degree of thermal deformation differs between the position corresponding to the gate 11 (also referred to as the gate position 11) and the general surface 12 (excluding the gate position 11), and the thickness distribution tends to differ as well. Therefore, when deforming the model 10, if morphing is performed across the entire model 10, there is a concern that the thickness near the gate 11 will not be constant when normalized. In order to solve this, the inventors of the embodiment have conducted diligent studies and have come to the conclusion that in the first morphing, the amount of thickness displacement is obtained across the entire model 10 and the first normalization (also referred to as first normalization) is performed, and in the second morphing, the gate position 11 is fixed within a predetermined range (for example, zero), and the second normalization (also referred to as second normalization) is performed using scaling from the first normalization.

[0061] (b) The plate thickness optimization method of the embodiment described above is characterized in that the resin member is formed by injecting resin from the gate 11 by injection molding, and the second shape deformation step is characterized in that when the model 10 is deformed, the amount of deformation at the position corresponding to the gate 11 in the model 10 is limited to a predetermined range, and a second normalization is performed based on the maximum and minimum values ​​of the displacement for each mesh 20 obtained in the first shape deformation step so that it becomes a predetermined plate thickness range, thereby generating the plurality of second learning samples.

[0062] The plate thickness optimization method of the embodiment described above, when configured as shown in (b) above, can keep the plate thickness deviation constant by considering the formability of the general surface 12 (excluding the area near the gate 11) and the area near the gate 11. Therefore, the plate thickness optimization method of this embodiment can optimize the plate thickness with high accuracy. That is, the plate thickness optimization method of the embodiment, when configured as shown in (b) above, can perform second normalization using the scaling of the first normalization, so that displacement amounts such as plate thickness can be compared and evaluated within the same scaling range. As a result, the plate thickness optimization method of the embodiment can generate analysis data regarding displacement amounts with high accuracy as a second training sample. Here, the range of deformation amount at the position corresponding to the gate 11 in the model 10 (also referred to as the gate position 11) can be limited to various ranges, but for example, it is good to fix it to zero. As a result, the plate thickness optimization method of the embodiment can obtain an optimal value of plate thickness that satisfies the target performance indicators (e.g., weight and heat resistance) from the gate 11 to the general surface 12 while keeping the plate thickness near the gate 11 constant.

[0063] (c) The plate thickness optimization method of the above-described embodiment is characterized by having an AI program that explores the relationship between the plate thickness of the resin member and the performance index, and by performing an AI learning step of training the AI ​​program with the training dataset, and an optimization search step of exploring the relationship between the plate thickness and the performance index based on the trained AI program.

[0064] The plate thickness optimization method of the embodiment described above can train an AI program using multiple second training samples and performance indicators as a training dataset, and explore the relationship between plate thickness and the performance indicators based on the trained AI program. Therefore, the plate thickness optimization method of the embodiment can perform efficient plate thickness optimization exploration using AI. As a result, the plate thickness optimization method of the embodiment can efficiently perform stable and accurate plate thickness design without relying on the experience or intuition of the designer, and thus a reduction in design man-hours can be expected.

[0065] (d) The plate thickness optimization method of the embodiment described above is characterized in that at least one of the first shape deformation step and the second shape deformation step deforms the model 10 based on design of experiments.

[0066] In the plate thickness optimization method of the embodiment described above, the model 10 is modified based on the design of experiments, so the first and second training samples can be efficiently generated without unnecessarily increasing the number of samples. As a result, the plate thickness optimization method of the embodiment can obtain accurate analysis results while speeding up the CAE analysis.

[0067] (e) The plate thickness optimization method of the above-described embodiment is characterized in that the performance index is based on one or a combination of any one of heat resistance, cold resistance, rigidity, fluidity, and impact resistance.

[0068] The plate thickness optimization method of the embodiment described above, when configured as shown in (e) above, can optimize plate thickness design based on performance indicators that affect plate thickness. Here, the performance indicators may include abrasion resistance, scratch resistance, paintability, thermal conductivity, and weather resistance.

[0069] Here, if the morphing deformation is not performed smoothly, there was a concern that the change in plate thickness for every 20 meshes would be stepwise, resulting in large changes in plate thickness.

[0070] (f) The plate thickness optimization method of the embodiment described above is characterized in that the deformation of the model 10 in the first shape deformation step and the second shape deformation step is performed by morphing deformation based on a spline function.

[0071] In the plate thickness optimization method of the embodiment described above, the deformation of Model 10 is performed by morphing deformation based on a spline function, so the change in the amount of plate thickness displacement for each mesh 20 can be changed along the spline curve. As a result, the plate thickness optimization method of the embodiment can smooth the change in plate thickness, and thus enable accurate optimization of plate thickness.

[0072] (g) In the plate thickness optimization method of the embodiment described above, the algorithm in the AI ​​program is characterized in that it is executed based on a graph neural network.

[0073] The plate thickness optimization method of the embodiment described above, when configured as shown in (g) above, can perform highly accurate plate thickness optimization using a graph neural network (GNN).

[0074] (h) In the plate thickness optimization method of the embodiment described above, the graph neural network is characterized in that it uses the adjacency matrix of each node 21 as an explanatory variable.

[0075] The plate thickness optimization method of the embodiment described above, when configured as shown in (h) above, can efficiently convert the relationships between multiple meshes 20 into explanatory variables.

[0076] (i) In the plate thickness optimization method of the embodiment described above, the graph neural network is characterized in that it uses a plate thickness value calculated by adding the plate thicknesses of a plurality of meshes 20 corresponding to each directly connected node 21 as an explanatory variable for each node 21.

[0077] The plate thickness optimization method of the embodiment described above can be configured as described in (i) above to efficiently obtain explanatory variables relating to the plate thickness values ​​between multiple meshes 20.

[0078] (j) The plate thickness optimization program of the above embodiment should be capable of executing (a) to (i) above in the information processing device.

[0079] Since the plate thickness optimization program of the above-described embodiment is executable on an information processing device, it can be executed on various information processing devices such as computers. Therefore, the plate thickness optimization program of the embodiment offers improved versatility.

[0080] The above describes the configuration and effects of the plate thickness optimization method and plate thickness optimization program of the present invention. However, the plate thickness optimization method and plate thickness optimization program of the present invention are not limited to the embodiments and modifications described above, and various modifications can be made within the scope of the present invention. For example, the plate thickness optimization method and plate thickness optimization program of the present invention may be as described in (a) above and can be applied to resin and other components of various shapes and sizes. Furthermore, the plate thickness optimization method of the present invention may, for example, not include some or all of the configurations related to (b) to (j) above, or may include some or all of (b) to (j) above and other configurations.

[0081] In this embodiment, the optimal plate thickness of a resin member is explored, but the present invention is not limited to this and can be used for members made of various materials other than resin members, for example. Also, in this embodiment, the mesh 20 is shown as being formed in a rectangular or triangular shape by modeling, but the shape of the mesh 20 can be of various shapes and sizes. Furthermore, the displacement amount and direction of displacement to be morphed in the first and second shape deformation steps can be changed to various displacement amounts and directions. In addition, the scaling in the first and second normalization is not limited to that of the embodiment, and various scaling methods can be used. Furthermore, modeling can be performed not only using the finite element method, but also by various other methods.

[0082] In this embodiment, the resin member is formed by injecting resin through the gate 11 by injection molding, but the invention is not limited to this, and the resin member can be formed by various manufacturing methods. In such cases, if a manufacturing method without a gate 11 is used, the area where the plate thickness is to be limited to a predetermined range (for example, fixed to zero) can be set as the position corresponding to the gate 11.

[0083] Furthermore, in this embodiment, at least one of the first shape deformation step and the second shape deformation step deforms the model 10 based on design of experiments, but the present invention is not limited thereto. For example, the plate thickness optimization method of the present invention may also be one that does not use design of experiments. In addition, the items such as plate thickness deformation amount and heat deformation temperature, and the displacement amount that are assigned in the design of experiments can be various items and displacement amounts depending on the target resin member, target value, etc.

[0084] In this embodiment, performance indicators are exemplified by one or a combination of any of the following: thermal deformation temperature, thermal deformation amount, stiffness, fluid analysis, and impact resistance. However, the performance indicators are not limited to these and various other indicators can be used.

[0085] In this embodiment, the deformation of Model 10 in the first and second shape deformation steps is performed by morphing deformation based on a spline function, but the present invention is not limited thereto. For example, either the first or second shape deformation step may be morphed deformation based on a spline function, or both the first and second shape deformation steps may be morphed deformation without using a spline function.

[0086] In this embodiment, the algorithm in the AI ​​program is executed based on a graph neural network (GNN), but the present invention is not limited to this, and the plate thickness optimization method of the present invention can perform analysis using algorithms other than graph neural networks. Furthermore, the plate thickness optimization method of the present invention does not only use an AI program, but also does not use AI. For example, the plate thickness optimization method of the present invention may use a graph neural network only for CAE analysis without using an AI program. Also, in this embodiment, the graph neural network uses the adjacency matrix of each node as an explanatory variable, but the present invention is not limited to this, and it may also not use the adjacency matrix. Furthermore, in this embodiment, the graph neural network uses the plate thickness value calculated by adding the plate thicknesses of multiple meshes 20 corresponding to each directly connected node as an explanatory variable for each node 21, but the present invention is not limited to this. For example, the plate thickness of node 21 at the boundary of each mesh 20 may be the average value of the plate thicknesses of adjacent meshes 20, etc.

[0087] In this embodiment, the program is provided as a plate thickness optimization program executable on an information processing device. However, the information processing device can be a server such as a cloud server, or various computing devices such as a personal computer or smartphone. Furthermore, the plate thickness optimization program may not only consist of a CAE and an AI program configured separately, but may also be integrated. In addition, the plate thickness optimization program may be divided into multiple modules. In such cases, each module may be deployed on multiple terminals or servers.

[0088] The above describes various embodiments and modifications of the plate thickness optimization method and plate thickness optimization program according to the present invention. However, the present invention is not limited to those exemplified in the embodiments and modifications described above, and it will be readily apparent to those skilled in the art that other embodiments may exist in the spirit and nature of the teachings, without departing from the scope of the claims. [Industrial applicability]

[0089] The present invention can be used in resin components of various vehicles. The present invention is particularly suitable for resin components such as the back doors of various vehicles. [Explanation of symbols]

[0090] 10: Model 11: Gate (Gate location) 12:General side 20: Mesh 20A: Mesh 20B: Mesh 21: Node 30: Morphing Box 31: Morphing Point

Claims

1. A method for optimizing the thickness of a resin component, which searches for the optimal thickness of the resin component, A first shape deformation step involves deforming a model of the resin member, which is divided into multiple meshes by modeling, to generate multiple first learning samples. A first normalization execution step in which a first normalization is performed on the plurality of first learning samples based on the displacement amount for each mesh, and the maximum and minimum values ​​of the displacement amount, A second shape deformation step generates a plurality of second learning samples so as to be within a predetermined plate thickness range, based on the first normalized value calculated in the first normalization execution step, A performance index analysis step in which performance indices corresponding to a predetermined amount of deformation are analyzed by performing structural analysis using the finite element method on the plurality of second learning samples, A training data generation step that generates the plurality of second training samples and the performance indicators as a training dataset, A plate thickness optimization method characterized by having the following features.

2. The aforementioned resin member is formed by injecting resin through a gate using injection molding. The second shape deformation step is, The plate thickness optimization method according to claim 1, characterized in that when deforming the model, the amount of deformation at the position corresponding to the gate in the model is limited to a predetermined range, and a plurality of second learning samples are generated by performing a second normalization based on the maximum and minimum values ​​of the displacement for each mesh obtained in the first shape deformation step so that the plate thickness falls within a predetermined range.

3. The system includes an AI program that explores the relationship between the thickness of the resin member and the performance index. An AI learning step in which the AI ​​program is trained using the training dataset, An optimization search step to explore the relationship between the plate thickness and the performance index based on a trained AI program, A plate thickness optimization method according to claim 1 or 2, characterized by performing the following:

4. A plate thickness optimization program according to claim 1 or 2 that can be executed in an information processing device.

5. The system includes an AI program that explores the relationship between the thickness of the resin member and the performance index. An AI learning step in which the AI ​​program is trained using the training dataset, An optimization search step to explore the relationship between the plate thickness and the performance index based on a trained AI program, A plate thickness optimization program according to claim 1 or 2, which can be executed in an information processing device.

Images