Three-dimensional shape data creation method, program, and three-dimensional shape data creation device

Abstract

A three-dimensional shape data creation method according to the present invention includes: an acquisition unit (11) acquiring first three-dimensional shape data for creating a target molded article; a conversion unit (12) converting the first three-dimensional shape data into second three-dimensional shape data that is a voxel set defined by at least one or more voxels, a basic point set defined by at least one or more points, or a point voxel set defined by at least one or more points and voxels; and a generation unit generating molded article three-dimensional shape data representing the shape of the target molded article, using a machine learning model trained in advance by machine learning with the second three-dimensional shape data as an input.

Description

Three-dimensional shape data creation method, program, and three-dimensional shape data creation apparatus

[0001] The present disclosure relates to a three-dimensional shape data creation method, a program, and a three-dimensional shape data creation apparatus.

[0002] For example, Patent Document 1 discloses a mold apparatus. This mold apparatus can injection-mold a molded product with a first mold, a second mold, and an insert.

[0003] Japanese Patent Application Laid-Open No. 2023-152777

[0004] For example, a producer may manufacture a molded product in response to a request from a customer. In this case, a developer receives input of data such as drawing data or three-dimensional data, and the customer's requirements including the appearance of the molded product, the arrangement position of the gate, locations where shape cannot be changed, material, and the like. In response to this input, the developer performs modeling of a preliminary draft through examination processes including imagining the mold structure for manufacturing the molded product, imagining the thinned shape, and imagining resin flow, then verifies formability risks for the modeled preliminary draft. Then, after verification, the developer proposes the molded product shape model to the customer.

[0005] However, since the examination process is performed through the developer's internal thinking, it may become dependent on specific individuals. Furthermore, since the developer performs modeling of the preliminary draft and verifies formability risks for the modeled preliminary draft, the developer's work cost increases.

[0006] In Patent Document 1, although a molded product is injection-molded with a first mold, a second mold, and an insert, no countermeasure is taken at all against suppressing the above-described dependence of the examination process on specific individuals and the increase in the developer's work cost.

[0007] An object of the present disclosure is to provide a three-dimensional shape data creation method and the like that can suppress the dependence of the examination process on specific individuals and the increase in the developer's work cost.

[0008] The method for creating three-dimensional shape data according to this disclosure includes: an acquisition unit acquiring first three-dimensional shape data for creating a target molded product; a conversion unit converting the first three-dimensional shape data into second three-dimensional shape data which is a set of voxels defined by at least one or more voxels, a set of basic points defined by at least one or more points, or a set of point voxels defined by at least one or more points and voxels; and a generation unit taking the second three-dimensional shape data as input and generating molded product three-dimensional shape data representing the shape of the target molded product using a machine learning model that has been previously trained by machine learning.

[0009] The method for creating three-dimensional shape data described herein can suppress the reliance on individual expertise in the design process and the increase in developer workload.

[0010] Figure 1 is a block diagram showing a three-dimensional shape data creation apparatus according to an embodiment. Figure 2 is a diagram showing the process from the first three-dimensional shape data to the second molded product three-dimensional shape data. Figure 3 is a diagram showing the training process of the first machine learning model. Figure 4 is a diagram showing the training process of the second machine learning model. Figure 5 is a flowchart showing an example of the operation of the three-dimensional shape data creation apparatus according to an embodiment.

[0011] The embodiments of this disclosure will be described in detail below with reference to the drawings. The embodiments described below are all specific examples of this disclosure. Therefore, the numerical values, shapes, materials, components, arrangement and connection configurations of components, steps, and the order of steps shown in the following embodiments are examples only and are not intended to limit this disclosure. Accordingly, any components in the following embodiments that are not described in an independent claim will be described as optional components.

[0012] Furthermore, each figure is a schematic diagram and not necessarily a strictly accurate representation. Therefore, for example, the scale and other aspects may not necessarily match in each figure. Also, in each figure, substantially identical components are given the same reference numerals, and redundant explanations are omitted or simplified.

[0013] The embodiments will be described in detail below with reference to the drawings.

[0014] (Embodiment) <Configuration and Function> First, the configuration and function of the three-dimensional shape data creation device 1 in this embodiment will be described with reference to Figures 1 to 4.

[0015] Figure 1 is a block diagram showing a three-dimensional shape data creation device 1 according to an embodiment. Figure 2 is a diagram showing the process from the first three-dimensional shape data to the second molded product three-dimensional shape data. Figure 2(a) shows the first three-dimensional shape data represented in STL format, Figure 2(b) shows the second three-dimensional shape data represented in voxel format, Figure 2(c) shows the first machine learning model, Figure 2(d) shows the first molded product three-dimensional shape data, Figure 2(e) shows the second machine learning model, and Figure 2(f) shows the second molded product three-dimensional shape data. Figure 3 is a diagram showing the learning process of the first machine learning model. Figure 4 is a diagram showing the learning process of the second machine learning model.

[0016] As shown in Figures 1 and 2, the three-dimensional shape data creation device 1 includes an acquisition unit 11, a conversion unit 12, a generation unit, and a storage unit.

[0017] The acquisition unit 11 is configured to acquire first three-dimensional shape data for creating a target molded product. The first three-dimensional shape data is data that represents the surface shape of the target (including contours and edges present on the surface) in three dimensions, and is a collection of coordinate data that expresses the positions of measurement points present on the surface of the target using an XYZ coordinate system. In this embodiment, the target is, for example, the shape of a flow path through which a substance moves inside the product. A flow path shape is a three-dimensional representation of the shape of a passage through which a liquid, gas, powder, or other substance flows. The first three-dimensional shape data shown in Figure 2(a) is an example of three-dimensional shape data in STL format that represents a flow path shape.

[0018] The conversion unit 12 is configured to convert the first three-dimensional shape data acquired by the acquisition unit 11 into second three-dimensional shape data. The second three-dimensional shape data is a set of voxels defined by at least one or more voxels, a set of basic points defined by at least one or more points, or a set of point voxels defined by at least one or more points and voxels. For example, as shown in Figure 2(b), the conversion unit 12 converts the first three-dimensional shape data in STL format representing the flow path shape into second three-dimensional shape data representing the flow path shape in voxel format.

[0019] The generation unit is configured to take second-dimensional shape data as input and generate three-dimensional shape data of a molded product that represents the shape of the target molded product using a pre-trained machine learning model that has been trained in advance through machine learning.

[0020] The machine learning model includes a first machine learning model and a second machine learning model. The molded product three-dimensional shape data includes first molded product three-dimensional shape data representing the shape of the first molded product, which is a preliminary stage to the target molded product, and second molded product three-dimensional shape data representing the shape of the second molded product, which is the target molded product.

[0021] Specifically, the generation unit has a first generation unit 21 and a second generation unit 22. The storage unit has a first storage unit 31 and a second storage unit 32.

[0022] As shown in Figures 1 and 2(b), (c), and (d), the first generation unit 21 is configured to take second three-dimensional shape data as input and generate first molded product three-dimensional shape data using a first machine learning model.

[0023] The first machine learning model is a machine learning model that takes into account the flow path shape when forming the first molded product, but it may also be a machine learning model that takes into account the wall thickness, ribs, and undercuts of the molded product wall that forms the flow path shape. Furthermore, the machine learning model may also be a machine learning model that takes into account fastening holes, and it may also be a machine learning model that takes into account the wall thickness, ribs, and undercuts of the molded product wall that forms the fastening holes. In other words, the first machine learning model may be a machine learning model that takes into account at least one of the flow path shape and fastening holes. A fastening hole is a hole through which fastening components such as bolts are passed and fastened to the fastening components.

[0024] The first molded product three-dimensional shape data is data that represents the surface shape of the first molded product (including the contour and edges present on the surface) in three dimensions, and is a collection of coordinate data that expresses the positions of measurement points present on the surface of the first molded product using an XYZ coordinate system. As shown in Figure 2(d), in the first molded product three-dimensional shape data of this embodiment, the first molded product having at least one of the flow channel shape and fastening holes is represented in three dimensions.

[0025] The first molded product is a preliminary molded product that represents the shape of the target molded product, and is a molded product in which at least a flow channel shape is formed by the molded product wall. The first molded product further has ribs and undercuts formed thereon. The first molded product does not have a gate that connects to a cavity for forming the target molded product, or a connecting part for connecting to the target molded product.

[0026] As shown in Figures 1 and 2(d), (e), and (f), the second generation unit 22 is configured to take the three-dimensional shape data of the first molded product as input and generate the three-dimensional shape data of the second molded product using a second machine learning model.

[0027] The second machine learning model is a machine learning model that takes into account the gates connected to the cavities for forming the target molded product (second molded product) when molding the target molded product (second molded product) using a mold. Furthermore, the second machine learning model may also take into account the connection parts for connecting to the second molded product when molding the target molded product (second molded product). The connection parts are, for example, connection parts for connecting to an object.

[0028] The second molded product three-dimensional shape data is data that represents the surface shape of the second molded product (including the contour and edges present on the surface) in three dimensions, and is a collection of coordinate data that expresses the positions of measurement points present on the surface of the second molded product using the XYZ coordinate system. As shown in Figure 2(f), in the second molded product three-dimensional shape data of this embodiment, the second molded product having at least one of the flow channel shape and fastening holes is represented in three dimensions.

[0029] The second molded product represents the shape of the target molded product and, in addition to the first molded product, has a gate formed thereon that connects to the cavity for forming the target molded product. The second molded product may also have a connecting portion formed thereon.

[0030] The second generation unit 22 outputs the generated three-dimensional shape data of the second molded product. For example, the three-dimensional shape data of the second molded product may be voxel-formatted three-dimensional shape data representing the shape of the product (second molded product).

[0031] The first storage unit 31 is composed of a semiconductor memory or the like, and stores the first machine learning model, the first molded product three-dimensional shape data, and the like. The first storage unit 31 may also be a semiconductor memory mounted on the first generation unit 21.

[0032] The second storage unit 32 is composed of a semiconductor memory or the like, and stores the second machine learning model, the second molded product three-dimensional shape data, and the like. The second storage unit 32 may also be a semiconductor memory mounted on the second generation unit 22.

[0033] Next, we will explain the training of the first machine learning model.

[0034] As shown in Figure 3, the first machine learning model is trained using the first learning unit 41a and the first identification unit 41b.

[0035] The first learning unit 41a is an architecture of a convolutional neural network such as U-Net. The first learning unit 41a consists of an encoder and a decoder, and can achieve high-resolution output while retaining three-dimensional shape data.

[0036] Furthermore, the first learning unit 41a is not limited to U-Net, but may be a CNN architecture applicable to encoder-decoder structures such as Attention U-Net, ResidualU-Net, SegNet, Tiramisu Network (FC-DenseNet), 3DU-Net, etc. In addition, other known architectures may be used in the first learning unit 41a.

[0037] The first learning unit 41a can take the second three-dimensional shape data as input data and generate first predicted three-dimensional shape data represented in voxel format using the first machine learning model under training.

[0038] The first identification unit 41b is a discriminator such as a Patch GAN used in image generation and transformation tasks.

[0039] The first identification unit 41b is not limited to Patch GANs, but may also be a local domain-focused discrimination architecture such as a Multi-Scale Discriminator, Dual Discriminator, or Patch-based Discriminators with Overlapping Patches. Furthermore, other known architectures may be used in the first identification unit 41b.

[0040] The first identification unit 41b receives the first predicted three-dimensional shape data output by the first learning unit 41a and the true value, which is the correct first molded product, as input data, and determines the truth value of the first predicted three-dimensional shape data. In other words, the first identification unit 41b compares the first predicted three-dimensional shape data with the true value and determines whether the first predicted three-dimensional shape data and the true value match. This first identification unit 41b is capable of evaluating the local quality of the output image such as the three-dimensional shape data, and can determine the truth value of each small patch that makes up the three-dimensional shape data and determine the truth value of the entire three-dimensional shape data by combining the results of the patches.

[0041] As a result, the first learning unit 41a can learn by calculating a loss function obtained from the result determined by the first identification unit 41b and updating the weights to minimize that loss. This update is performed, for example, using backpropagation and based on an optimization algorithm such as SGD (Stochastic Gradient Descent). Therefore, the first learning unit 41a can generate first predicted three-dimensional shape data that is identical to the true value.

[0042] For example, in the first learning unit 41a of this embodiment, a first machine learning model is trained so that the flow path shape and the wall thickness, ribs, and undercuts of the molded product wall that forms the flow path shape can be taken into consideration.

[0043] For example, the training of the first machine learning model includes conditions for setting the molded product wall portion for forming the flow channel shape and conditions for setting the wall thickness of the molded product wall portion. Specifically, the first learning unit 41a and the first identification unit 41b can incorporate constraints such as the main functional parts of the flow channel shape, the appearance requirements of the molded product, and the location information of parts whose shape cannot be changed into the loss function. Therefore, the first learning unit 41a can train the first machine learning model while taking these constraints into consideration.

[0044] For example, in the training of the first machine learning model, the position where the rib is to be left can be further added as a constraint. Specifically, in the first training unit 41a and the first discrimination unit 41b, the position where the rib is to be left (rib position information) can be incorporated into the loss function as a constraint. Therefore, the first training unit 41a can train the first machine learning model in consideration of the constraint.

[0045] As described above, the first generation unit 21 can generate the first molded product three-dimensional shape data based on the input second three-dimensional shape data by using the first machine learning model trained by the first training unit 41a.

[0046] Next, training of the second machine learning model will be described.

[0047] As shown in Fig. 4, in the training of the second machine learning model, a second training unit 42a and a second discrimination unit 42b are used.

[0048] The second training unit 42a adopts a convolutional neural network architecture such as U-Net. The second training unit 42a is composed of an encoder, a decoder and the like, and can achieve high-resolution output while retaining three-dimensional shape data.

[0049] Note that the second training unit 42a is not limited to U-Net, and may be a CNN architecture applicable to encoder-decoder structures such as Attention U-Net, Residual U-Net, SegNet, Tiramisu Network (FC-DenseNet), and 3D U-Net, for example. Also, other known architectures may be used in the second training unit 42a.

[0050] The second training unit 42a takes the first molded product three-dimensional shape data as input data, and can generate second predicted three-dimensional shape data expressed in a voxel format using the second machine learning model under training.

[0051] The second discrimination unit 42b is a discriminator such as Patch GAN used in image generation and conversion tasks.

[0052] Furthermore, the second identification unit 42b is not limited to Patch GANs, but may be a local domain-focused discrimination architecture such as a Multi-Scale Discriminator, Dual Discriminator, or Patch-based Discriminators with Overlapping Patches. In addition, other known architectures may be used in the second identification unit 42b.

[0053] The second identification unit 42b receives the second predicted three-dimensional shape data output by the second learning unit 42a and the true value, which is the correct second molded product, as input data, and determines the truth value of the second predicted three-dimensional shape data. In other words, the second identification unit 42b compares the second predicted three-dimensional shape data with the true value and determines whether the second predicted three-dimensional shape data and the true value match. This second identification unit 42b is capable of evaluating the local quality of the output image such as three-dimensional shape data, and can determine the truth value of each small patch that makes up the three-dimensional shape data and determine the truth value of the entire three-dimensional shape data by combining the results of the patches.

[0054] As a result, the second learning unit 42a can learn by calculating the loss function obtained from the result determined by the second discrimination unit 42b and updating the weights to minimize that loss. This update is performed using, for example, backpropagation and based on an optimization algorithm such as SGD (Stochastic Gradient Descent). Therefore, the second learning unit 42a can generate second predicted three-dimensional shape data that is identical to the true value.

[0055] For example, in the second learning unit 42a of this embodiment, the gate position is generated while considering the opening and closing parts of the mold. The second learning unit 42a trains a second machine learning model so that when molding a target molded product using a mold, it can consider the gate connected to the cavity for forming the target molded product. Furthermore, the second learning unit 42a trains a second machine learning model so that when molding a target molded product, it can consider the connection parts for connecting to the target molded product.

[0056] For example, in training the second machine learning model, a condition is added regarding locations where gates cannot be placed. Specifically, the second learning unit 42a and the second discrimination unit 42b can incorporate locations where gates cannot be placed (gate placement impossible location information) as constraints into the loss function. Therefore, the second learning unit 42a can train the second machine learning model while taking these constraints into consideration.

[0057] For example, in training the second machine learning model, additional constraints can be added, such as locations where connection parts cannot be placed. Specifically, the second learning unit 42a and the second discrimination unit 42b can incorporate locations where connection parts cannot be placed (connection part placement impossible location information) as constraints into the loss function. Therefore, the second learning unit 42a can train the second machine learning model while taking these constraints into consideration.

[0058] In this way, the second generation unit 22 can generate second molded product three-dimensional shape data based on the input first molded product three-dimensional shape data by using the second machine learning model learned in the second learning unit 42a.

[0059] Such a first machine learning model may take into account the thickness of the fastening holes and the wall portion of the molded product that forms the fastening holes. For example, after the first generation unit 21 generates the first molded product three-dimensional shape data using the first machine learning model, the developer may set the positions of the fastening holes in the first molded product three-dimensional shape data, and the first molded product three-dimensional shape data including the set fastening holes may be input to the second generation unit 22. In this case, the second generation unit 22 generates the second molded product three-dimensional shape data based on the first molded product three-dimensional shape data in which the fastening holes are set. Alternatively, the conversion unit 12 may convert the first three-dimensional shape data including the fastening holes set by the developer into second three-dimensional shape data, and the first generation unit 21 may generate the first molded product three-dimensional shape data that takes the fastening holes into consideration. The developer can check the generated first molded product three-dimensional shape data, and if there is a problem with the arrangement of the fastening holes, they can correct the positions of the fastening holes and generate the first molded product three-dimensional shape data again.

[0060] In the above description, the flow path shape was used as an example of the first three-dimensional shape data, but it is not limited to this. The first three-dimensional shape data may also be the internal structure of a molded product other than the flow path shape. For example, the first three-dimensional shape data may be the internal structure of a nut, the internal structure of a housing such as a case, the internal structure of a container, and the internal structure of a cap. Therefore, the machine learning model of this embodiment can be applied not only to the flow path shape but also to these internal structures of molded products. In other words, the first machine learning model may be a machine learning model that takes into account at least one of the flow path shape, fastening holes, and the internal structure of a molded product.

[0061] For example, the first machine learning model of this embodiment can also be applied to molded products having internal structures other than flow channel shapes, such as nuts, housings, containers, or caps.

[0062] Specifically, the first machine learning model may further incorporate models that take into account internal structures such as nuts, housings, containers, or caps, by training with training data of these internal structures. The training method is the same as in the case of the flow path shape described above, in which the first machine learning model is trained by replacing the training data of the flow path shape with the training data of these internal structures.

[0063] Here, the internal structure of nuts, housings, containers, or caps refers to, for example, the shape of the internal space of the molded product, the size of the internal space, the arrangement of the internal space, internal partitions, support structures, functional members, reinforcing parts, steps, and protrusions. Note that the internal structure of nuts, housings, containers, or caps excludes the fastening holes mentioned above.

[0064] For example, by obtaining second three-dimensional shape data from first three-dimensional shape data in the same manner as described above, when the second three-dimensional shape data showing the internal structure is input to the first machine learning model that takes the internal structure into consideration, the first generation unit 21 generates first molded product three-dimensional shape data in which the internal structure is formed. If this first molded product three-dimensional shape data represents the final molded product, then data showing the target molded product (nut, housing, container, or cap, etc.) is obtained.

[0065] Furthermore, the first machine learning model that takes the internal structure into consideration may also take into account the wall thickness, ribs, and undercuts of the molded product. The learning method is the same as in the case of the flow path shape described above, and the first machine learning model is trained by replacing the learning data for the flow path shape with learning data for the wall thickness, ribs, and undercuts of the molded product. As a result, the first generation unit 21 can generate three-dimensional shape data of the first molded product that takes into account not only the inside of the case or container, but also the wall thickness, ribs, and undercuts of the molded product.

[0066] Furthermore, the second machine learning model takes the gate position into consideration by learning using training data of the gate positions. The learning method is the same as that used for training the second machine learning model described above. Therefore, the second generation unit 22 takes the first molded product three-dimensional shape data as input and generates second molded product three-dimensional shape data using the second machine learning model. This makes it possible to obtain data that represents the target molded product.

[0067] In particular, by incorporating wall thickness, rib position, undercut position, and gate position as constraints into the loss function, it is expected that the first or second molded product three-dimensional shape data will be generated that takes into account appropriate wall thickness, rib position, undercut position, and gate position. In this way, the machine learning model is expected to be applicable to the design of a wide range of molded products, not only molded products with flow channel shapes, but also various molded products such as nuts, housings, containers, and caps, by learning their respective internal shapes and constraints.

[0068] <Operation> First, the operation of the three-dimensional shape data creation device 1 in this embodiment will be explained with reference to Figure 5.

[0069] Figure 5 is a flowchart showing an example of the operation of the three-dimensional shape data creation device 1 according to the embodiment.

[0070] The developer extracts first three-dimensional shape data from product data such as drawing data or 3D data at the request of the customer. In other words, the developer generates first three-dimensional shape data by extracting essential shapes such as the flow path shape and the internal structure of the molded product other than the flow path shape, based on the customer's product data. By inputting the extracted first three-dimensional shape data into the 3D shape data creation device 1, the acquisition unit 11 can acquire the first three-dimensional shape data (S11).

[0071] Next, the conversion unit 12 converts the first three-dimensional shape data acquired by the acquisition unit 11 into second three-dimensional shape data (S12). For example, the conversion unit 12 converts the first three-dimensional shape data into second three-dimensional shape data which is a set of voxels defined by at least one or more voxels, a set of basic points defined by at least one or more points, or a set of point voxels defined by at least one or more points and voxels, so that the first machine learning model can easily generate the first molded product three-dimensional shape data.

[0072] Next, the first generation unit 21 takes the second three-dimensional shape data as input and generates the first molded product three-dimensional shape data using the first machine learning model (S13). For example, the first generation unit 21 uses the first machine learning model to generate the first molded product three-dimensional shape data based on the second three-dimensional shape data in voxel format, taking undercuts into consideration and forming the flow channel shape, the internal structure of the molded product other than the flow channel shape, etc.

[0073] Next, the second generation unit 22 takes the three-dimensional shape data of the first molded product as input and generates the three-dimensional shape data of the second molded product using the second machine learning model (S14). For example, the second generation unit 22 uses the first machine learning model to generate the three-dimensional shape data of the second molded product, which has gates formed that are connected to cavities for forming the target molded product, based on the three-dimensional shape data of the first molded product in voxel format. In this way, the three-dimensional shape data of the second molded product representing the target molded product (second molded product) can be obtained.

[0074] <Effects and Effects> Next, the effects and effects of the three-dimensional shape data creation method, program, and three-dimensional shape data creation apparatus 1 in this embodiment will be described.

[0075] The three-dimensional shape data creation method of Technology 1 in this embodiment includes: an acquisition unit 11 acquiring first three-dimensional shape data for creating a target molded product; a conversion unit 12 converting the first three-dimensional shape data into second three-dimensional shape data which is a set of voxels defined by at least one or more voxels, a set of basic points defined by at least one or more points, or a set of point voxels defined by at least one or more points and voxels; and a generation unit taking the second three-dimensional shape data as input and generating molded product three-dimensional shape data representing the shape of the target molded product using a machine learning model that has been previously trained by machine learning.

[0076] According to this, by inputting second-dimensional shape data into the generation unit, a machine learning model can generate three-dimensional shape data for a molded product.

[0077] Therefore, it is possible to suppress the personalization of the review process and the increase in the workload of developers.

[0078] The three-dimensional shape data creation method of Technology 2 in this embodiment is the three-dimensional shape data creation method described in Technology 1. In this case, the first three-dimensional shape data is data of a molded product having at least one of the following: a flow path shape for the movement of a material, fastening holes for passing fastening components, and an internal structure of the molded product.

[0079] According to this method, three-dimensional shape data of molded products can be generated using machine learning models for any type of molded product, including those with flow channel shapes, fastening holes, and internal structures such as housings and containers. Therefore, it is possible to suppress the reliance on individual expertise in the design process and the increase in developer workload for both flow channel shapes and various other types of molded products.

[0080] Furthermore, since the present invention can similarly generate three-dimensional shape data for molded products that combine multiple elements of flow path shape, fastening holes, and internal structure, it can also be applied to the design of molded products with more complex structures.

[0081] The three-dimensional shape data creation method of Technology 3 in this embodiment is the three-dimensional shape data creation method described in Technology 1 or 2. In this case, the machine learning model includes a first machine learning model and a second machine learning model, the molded product three-dimensional shape data includes first molded product three-dimensional shape data which is a preliminary stage to the shape representing the target molded product and second molded product three-dimensional shape data which represents the shape of the target molded product, the generation unit includes a first generation unit 21 and a second generation unit 22, the first generation unit 21 takes the second three-dimensional shape data as input and generates first molded product three-dimensional shape data using the first machine learning model, and the second generation unit 22 takes the first molded product three-dimensional shape data as input and generates second molded product three-dimensional shape data using the second machine learning model.

[0082] According to this method, since the three-dimensional shape data for the second molded product, which will become the final product, can be generated after generating the three-dimensional shape data for the first molded product (which is the preliminary stage), the increase in processing load applied to the generation unit can be suppressed compared to generating the three-dimensional shape data for the second molded product all at once.

[0083] For example, if a gate is included in the three-dimensional shape data of a molded product, the model may still consider it correct even if the gate is located in a different position. When multiple correct candidate models exist, even if the model generates the three-dimensional shape data of a molded product using a machine learning model, the accuracy of the model may decrease. However, in this embodiment, by generating the first three-dimensional shape data of a molded product, which is a preliminary stage to the target molded product, and then generating the second three-dimensional shape data of a molded product, it is possible to suppress the decrease in accuracy of the second three-dimensional shape data of a molded product that represents the target molded product.

[0084] The three-dimensional shape data creation method of Technology 4 in this embodiment is the three-dimensional shape data creation method described in Technology 2. In this case, the molded product three-dimensional shape data is data of a target molded product having at least one of the following: flow path shape, fastening holes, and internal structure.

[0085] According to this method, three-dimensional shape data of a molded product in which at least one of the following is formed: flow channel shape, fastening holes, and internal structure can be obtained.

[0086] The three-dimensional shape data creation method of Technology 5 in this embodiment is the three-dimensional shape data creation method described in Technology 4. In this case, the machine learning model is a machine learning model that, when molding the target molded product, takes into account at least one of the flow path shape, fastening holes, and internal structure, as well as the wall thickness of the molded product wall, ribs, and undercuts, and also takes into account the gate connected to the cavity for forming the target molded product using a mold.

[0087] According to this, a single machine learning model can generate three-dimensional shape data for a molded product that takes into account at least one of the following: flow path shape, fastening holes, and internal structure, as well as the wall thickness of the molded product, ribs, undercuts, and gates connected to the cavity. Therefore, it is expected that a single three-dimensional shape data for a molded product can be obtained for molding the target product.

[0088] The three-dimensional shape data creation method of Technology 6 in this embodiment is the three-dimensional shape data creation method described in Technology 3. In this case, the first molded product three-dimensional shape data is data of a first molded product having at least one of the following: a flow channel shape, fastening holes for passing fastening components, and an internal structure of the molded product; the first machine learning model is a machine learning model that takes into account at least one of the flow channel shape, fastening holes, and internal structure, the wall thickness of the molded product wall portion forming the flow channel shape, ribs, and undercuts; and the second machine learning model is a machine learning model that takes into account the gate connected to the cavity for forming the molded product when molding a target molded product using a mold.

[0089] According to this, the first machine learning model can generate three-dimensional shape data of a first molded product that takes into account at least one of the flow path shape, fastening holes, and internal structure, as well as the wall thickness of the molded product, ribs, and undercuts. Therefore, it is expected that a second three-dimensional shape data for molding the target molded product can be obtained.

[0090] Furthermore, the second machine learning model enables the generation of a second molded part's three-dimensional shape data that takes the gate into account. Therefore, it becomes possible to obtain the second molded part's three-dimensional shape data necessary for molding the target product.

[0091] The three-dimensional shape data creation method of Technology 7 in this embodiment is the three-dimensional shape data creation method described in Technology 6. In this case, the second machine learning model is a machine learning model that further takes into account the connection part for connecting to the target molded product when molding the target molded product.

[0092] According to this, the second machine learning model will be able to generate three-dimensional shape data of a second molded product that takes gates and connections into account. Therefore, it is expected that we can obtain three-dimensional shape data of a second molded product necessary for molding the target molded product.

[0093] The program of technology 8 in this embodiment is a program that allows a computer to execute the three-dimensional shape data creation method described in any one of technologies 1 to 7.

[0094] This program also produces the same effects as described above.

[0095] The three-dimensional shape data creation apparatus 1 of technology 9 in this embodiment includes: an acquisition unit 11 that acquires first three-dimensional shape data for creating a target molded product; a conversion unit 12 that converts the first three-dimensional shape data into second three-dimensional shape data which is a set of voxels defined by at least one or more voxels, a set of basic points defined by at least one or more points, or a set of point voxels defined by at least one or more points and voxels; and a generation unit that takes the second three-dimensional shape data as input and generates molded product three-dimensional shape data representing the shape of the target molded product using a machine learning model that has been previously learned by machine learning.

[0096] This three-dimensional shape data creation device 1 also produces the same effects as described above.

[0097] (Other Modifications) The three-dimensional shape data creation method and the like related to this disclosure have been described above based on the embodiments described above, but this disclosure is not limited to these embodiments. Various modifications to the embodiments that a person skilled in the art can conceive of may also be included within the scope of this disclosure, as long as they do not deviate from the spirit of this disclosure.

[0098] For example, in the method for creating three-dimensional shape data according to this disclosure, the generation unit may directly generate second molded product three-dimensional shape data from second three-dimensional shape data using a single machine learning model. In other words, a single generation unit may generate molded product three-dimensional shape data corresponding to second molded product three-dimensional shape data representing the shape of the target molded product using a single machine learning model.

[0099] In this case, the machine learning model may be one that takes into account the flow path shape, the wall thickness of the molded product wall forming the flow path shape, ribs and undercuts, and the gate connected to the cavity for forming the target molded product using a mold. In this case, the generation unit may be configured to take second three-dimensional shape data as input and generate three-dimensional molded product shape data in voxel format representing the shape of the product (target molded product) using the machine learning model.

[0100] Furthermore, while the three-dimensional shape data creation method described herein uses molded products having a flow path shape (e.g., valve bodies, fitting pipes) as examples, it is not limited to these. For example, the molded products may include furniture such as chairs, desks, and tables; tableware such as cups, knives, and forks; tools such as wrenches and screwdrivers; and mechanical parts such as gears.

[0101] Furthermore, the generation units and the like used in the three-dimensional shape data creation method and the like related to this disclosure are typically implemented as LSIs, which are integrated circuits. These may be individually integrated into a single chip, or some or all of them may be integrated into a single chip.

[0102] Furthermore, integrated circuit implementation is not limited to LSIs; it may also be achieved using dedicated circuits or general-purpose processors. Alternatively, an FPGA (Field Programmable Gate Array), which can be programmed after LSI manufacturing, or a reconfigurable processor that allows for the reconfiguration of the connections and settings of the circuit cells within the LSI, may be used.

[0103] In each of the above embodiments, each component may be implemented by dedicated hardware or by executing a software program suitable for each component. Each component may also be implemented by a program execution unit such as a CPU or processor reading and executing a software program recorded on a recording medium such as a hard disk or semiconductor memory.

[0104] Furthermore, all figures used above are illustrative to illustrate the present disclosure, and the embodiments of this disclosure are not limited to the figures exemplified.

[0105] Furthermore, the division of functional blocks in the block diagram is just one example; multiple functional blocks can be implemented as a single functional block, a single functional block can be divided into multiple parts, or some functions can be moved to other functional blocks. In addition, the functions of multiple functional blocks with similar functions can be processed in parallel or time-sharing by a single piece of hardware or software.

[0106] Furthermore, the order in which each step in the flowchart is performed is illustrative for the purpose of specifically illustrating this disclosure, and may be in a different order. Also, some of the above steps may be performed simultaneously (in parallel) with other steps.

[0107] Furthermore, this disclosure also includes forms that can be obtained by applying various modifications to the above embodiments that a person skilled in the art could conceive, as well as forms that can be realized by arbitrarily combining the components and functions of the embodiments without departing from the spirit of this disclosure.

[0108] This disclosure is applicable to the manufacture of products using molds.

[0109] 1. Three-dimensional shape data creation device 11. Acquisition unit 12. Conversion unit 21. First generation unit (generation unit) 22. Second generation unit (generation unit)

Claims

1. A method for creating three-dimensional shape data, comprising: an acquisition unit acquiring first three-dimensional shape data for creating a target molded product; a conversion unit converting the first three-dimensional shape data into second three-dimensional shape data which is a set of voxels defined by at least one or more voxels, a set of basic points defined by at least one or more points, or a set of point voxels defined by at least one or more points and voxels; and a generation unit generating molded product three-dimensional shape data representing the shape of the target molded product using a machine learning model that has been previously trained by machine learning, with the second three-dimensional shape data as input.

2. The method for creating three-dimensional shape data according to claim 1, wherein the first three-dimensional shape data is data of a molded product having at least one of the following: a flow path shape of the molded product for the movement of a substance, fastening holes for passing fastening components, and an internal structure of the molded product.

3. The method for creating three-dimensional shape data according to claim 1, wherein the machine learning model includes a first machine learning model and a second machine learning model, the molded product three-dimensional shape data includes first molded product three-dimensional shape data which is a preliminary stage to the shape representing the target molded product and second molded product three-dimensional shape data which represents the shape of the target molded product, the generation unit includes a first generation unit and a second generation unit, the first generation unit takes the second three-dimensional shape data as input and generates the first molded product three-dimensional shape data using the first machine learning model, and the second generation unit takes the first molded product three-dimensional shape data as input and generates the second molded product three-dimensional shape data using the second machine learning model.

4. The method for creating three-dimensional shape data according to claim 2, wherein the three-dimensional shape data of the molded product is data representing a target molded product having at least one of the flow channel shape, the fastening hole, and the internal structure.

5. The method for creating three-dimensional shape data according to claim 4, wherein the machine learning model is a machine learning model that, when forming the target molded product, takes into account at least one of the flow path shape, the fastening hole, and the internal structure, the wall thickness of the molded product wall, ribs, and undercuts, and also takes into account the gate connected to the cavity for forming the target molded product using a mold.

6. The method for creating three-dimensional shape data according to claim 3, wherein the first molded product three-dimensional shape data is data of a first molded product having at least one of the flow channel shape, fastening holes for passing fastening components, and internal structure of the molded product; the first machine learning model is a machine learning model that takes into account at least one of the flow channel shape, fastening holes, and internal structure, the wall thickness of the molded product wall portion forming the flow channel shape, ribs, and undercuts; and the second machine learning model is a machine learning model that takes into account the gate connected to the cavity for forming the target molded product when molding the target molded product using a mold.

7. The method for creating three-dimensional shape data according to claim 6, wherein the second machine learning model is a machine learning model that further takes into account a connection portion for connecting to the target molded product when molding the target molded product.

8. A computer program capable of executing the method for creating three-dimensional shape data according to any one of claims 1 to 7.

9. A three-dimensional shape data creation device comprising: an acquisition unit that acquires first three-dimensional shape data for creating a target molded product; a conversion unit that converts the first three-dimensional shape data into second three-dimensional shape data which is a set of voxels defined by at least one or more voxels, a set of basic points defined by at least one or more points, or a set of point voxels defined by at least one or more points and voxels; and a generation unit that takes the second three-dimensional shape data as input and generates molded product three-dimensional shape data representing the shape of the target molded product using a machine learning model that has been previously trained by machine learning.

Images