Design device and design method

Abstract

To provide a design device capable of generating new shape data corresponding to a designated new shape code other than a teacher shape code.SOLUTION: In a design device 1, an encoder unit 3 receives shape data 2 and a shape code 7, which is a value corresponding to the shape data 2, calculates feature information of the shape data 2, and stores feature group information 4, which is information associating the feature information with the shape code 7. A decoder unit 5 generates new shape data 6 corresponding to a new shape code 8 from the feature group information 4. As a result, the new shape data 6 corresponding to the designated new shape code 8 other than a teacher shape code is generated.SELECTED DRAWING: Figure 1

Description

[Technical field]

[0001] The present invention relates to a design device and a design method for a shape. [Background technology]

[0002] Traditionally, mechanical design has generally been based on rule-based design methods based on engineering knowledge and unique design standards, and these methods have produced significant results in the field of mechanical design.

[0003] On the other hand, this kind of rule-based design method is a mass of know-how, and requires the skills of an expert, not only in how to use the design tools but also in how to apply the design standards. In today's world, where it is difficult to train people who are thoroughly familiar with every aspect of the design, passing on know-how has become a major issue.

[0004] Therefore, we focus on feature group information extraction technology that uses convolutional neural networks, which is one of the machine learning techniques.

[0005] In conventional rule-based design methods, humans define design patterns to be extracted from data, but deep learning using convolutional neural networks can do this automatically.

[0006] This makes it possible to establish a design process based on objective judgment, rather than relying on the know-how of experts as in the past.

[0007] Patent Document 1 describes a technique in which a convolutional neural network learns training shape data, extracts feature group information, inputs a target performance value, and then restores the same shape data. [Prior art documents] [Patent documents]

[0008] [Patent Document 1] JP 2020-173701 A Summary of the Invention [Problem to be solved by the invention]

[0009] However, in the conventional technology described in Patent Document 1, training shape data is trained on a convolutional neural network, making it difficult to generate new shape data other than the training data.

[0010] An object of the present invention is to provide a designing apparatus and a designing method capable of generating new shape data corresponding to a designated shape code other than a teaching shape code. [Means for solving the problem]

[0011] In order to achieve the above object, the present invention is configured as follows.

[0012] The design device includes an encoder unit that receives shape data and a shape code that is a value corresponding to the shape data, calculates feature information for the shape data, and stores feature group information that is information that associates the feature information with the shape code, and a decoder unit that generates new shape data that corresponds to the new shape code from the feature group information.

[0013] In addition, in the design method, shape data and a shape code which is a value corresponding to the shape data are received, feature information of the shape data is calculated, feature group information which is information that associates the feature information with the shape code is stored, and new shape data corresponding to the new shape code is generated from the stored feature group information. Effect of the Invention

[0014] According to the present invention, it is possible to provide a designing device and a designing method capable of generating new shape data corresponding to a specified new shape code other than a teaching shape code. [Brief description of the drawings]

[0015] [Figure 1] FIG. 1 is a diagram illustrating a design device that executes an algorithm according to a first embodiment. [Diagram 2] 4 is a flowchart showing a process of an encoder unit in the first embodiment. [Diagram 3] 4 is a flowchart showing a process of a decoder unit in the first embodiment. [Figure 4] FIG. 11 is a diagram showing an example of a shape data group that is training data. [Diagram 5] FIG. 3 is a diagram showing an example of shape data obtained by the design device of the first embodiment. [Figure 6] 13 is a flowchart showing a process of a decoder unit in the second embodiment. [Figure 7] FIG. 1 is a diagram showing a design device having a display unit. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0016] Hereinafter, the present invention will be described in detail based on the embodiments shown in the accompanying drawings. EXAMPLES

[0017] Example 1 A first embodiment of the present invention will be described.

[0018] FIG. 1 is a diagram showing a design device 1 that executes an algorithm according to a first embodiment of the present invention.

[0019] In FIG. 1, a design device 1 according to the first embodiment includes a shape data group 2, a shape code group 7, an encoder unit 3, and a decoder unit 5.

[0020] The encoder unit 3 receives a shape data group 2 which is teacher data and a shape code group 7 which is a value corresponding to each shape in the shape data group 2, calculates feature information of the shape data group 2, and stores feature group information 4 which is information associating the feature information with the shape code group 7. The decoder unit 5 generates new shape data 6 which corresponds to a new shape code 8 from the feature group information 4.

[0021] In addition, the encoder unit 3 and the decoder unit 5 can be used independently as a feature extractor and a shape data generator, respectively. In the first embodiment, the encoder unit 3 and the decoder unit 5 have a neural network structure using a convolution layer, and can quickly extract feature group information 4 from a large number of shape data groups 2.

[0022] First, a user inputs a shape data group 2, which is teaching data, and a corresponding shape code group 7 to a design device 1 having a set algorithm.

[0023] Next, the shape data group 2 and the corresponding shape code group 7 are temporarily stored in the encoder unit 3 as feature group information 4, which is information associating the shape data group 2 with the shape code group 7. Furthermore, when the user inputs a new shape code 8 to be generated to the design device 1, new shape data 6 corresponding to the new shape code 8 to be generated is constructed in the decoder unit 5 and output from the design device 1.

[0024] Fig. 2 is a flowchart showing the process of the encoder unit 3 in the first embodiment of the present invention. Also, Fig. 3 is a flowchart showing the process of the decoder unit 5 in the first embodiment of the present invention.

[0025] 2, a shape data group 2 and a shape code group 7 corresponding to the shape data group 2 are input to an encoder unit 3 of a design device 1 (step S1). Next, the encoder unit 3 generates feature group information 4 based on the shape data group 2 and the shape code group 7.

[0026] In FIG. 3, the decoder unit 5 converts the new shape code 8 to be generated into a one-hot vector (step S10).

[0027] Next, the decoder unit 5 inputs the one-hot vector to the design device 1 (step S11).

[0028] Then, in a step (S12), the decoder unit 5 generates new shape data 6 by combining the feature group information 4. This new shape data 6 is data that includes data different from the shape data group 2 that is the teacher data.

[0029] Then, the decoder unit 5 outputs the new shape data 6 (step S13).

[0030] In the first embodiment, the decoder unit 5 generates an image by synthesizing feature group information 4 corresponding to a plurality of shape codes, thereby making it possible to generate new shape data 6 that is different from the shape data group 2 that is the teacher data.

[0031] To synthesize the feature group information 4, the ratio of feature vectors input when constructing data in the decoder unit 5 is changed. For example, when shape codes of n=1, 2, 3, 4 are given, the shape code information to be generated is converted to a one-hot vector and input, such as [1 0 0 0] for n=1 and [0 1 0 0] for n=2, but this can be changed according to the desired ratio of synthesis, for example, to [0.5 0.5 0 0] if the desired ratio of synthesis is 50%.

[0032] The above example, [0.5 0.5 0 0], is how to give a one-hot vector when you want to output n=1.5. The feature information for n=1 and n=2 is combined at 50% each in the decoder unit 5 to obtain new shape data 6.

[0033] FIG. 4 is a diagram showing an example of a shape data group that is training data, and FIG. 5 is a diagram showing an example of shape data obtained by the design device 1 of the first embodiment.

[0034] The shape data shown in Fig. 5 is obtained by adding new shape data different from the teacher data to the teacher data shown in Fig. 4. The new shape data is shape data corresponding to a shape value between a certain shape data and the next shape data in the shape data group that is the teacher data. The explanation will be given assuming that the synthesis rate is 50%.

[0035] The shape data group, which is the training data shown in Figure 4, is a shape data group consisting of data (A) in which a rectangular component is inclined at 25 degrees from the horizontal, data (B) in which a rectangular component is inclined at 0 degrees from the horizontal, and data (C) in which a rectangular component is inclined at -25 degrees from the horizontal.

[0036] The shape data shown in Figure 5 is data (AB) which is a 50% combination of data (A) and data (B) from Figure 4 and tilted at 12.5 degrees from the horizontal, and data (BC) which is a 50% combination of data (B) and data (C) from Figure 4 and tilted at -12.5 degrees from the horizontal, added to data (A), (B) and (C).

[0037] The new shape data 6 corresponds to the data (AB) and data (BC) shown in FIG.

[0038] In this way, according to the first embodiment of the present invention, by synthesizing the training data, new shape data 6 can be obtained by adding shape data that is not included in the training data group.

[0039] As described above, according to the first embodiment of the present invention, a design device and a design method can be provided that can cause an encoder unit 3 having a neural network structure to learn a shape data group 2 and a shape code group 7 which is a value corresponding to the shape data group 2, calculate feature information of the shape data, extract feature group information 4 which is information that associates this feature information with the shape code, and generate new shape data 6 corresponding to a specified new shape code 8 other than the teacher shape code by a decoder unit 5 based on the extracted feature group information 4.

[0040] Furthermore, according to the first embodiment of the present invention, it is possible to provide a simple design device and a design method that does not utilize a design technique that is dependent on individual skills based on fluid engineering knowledge and design standards.

[0041] Example 2 Next, a second embodiment of the present invention will be described.

[0042] The overall configuration of the second embodiment is similar to that of the first embodiment, but the functions of the encoder unit 3 and the decoder unit 5 of the second embodiment are different from those of the first embodiment. Since the overall configuration of the second embodiment is similar to that shown in FIG. 1, the illustration of the overall configuration of the second embodiment is omitted.

[0043] The design device 1 of the second embodiment is equipped with an encoder unit 3 that receives a shape data group 2 and a shape code group 7 which are values ​​corresponding to each shape in the shape data group 2, calculates feature information of the shape data group 2, and stores feature group information 4 which is information that associates the feature information with the shape code group 7, and a decoder unit 5 that generates new shape data 6 corresponding to a new shape code 8 from the feature group information 4.

[0044] Moreover, the encoder unit 3 and the decoder unit 5 can be used independently as a feature extractor and a shape data generator, respectively.

[0045] In the second embodiment, the encoder unit 3 and the decoder unit 5 have a neural network structure using a convolution layer, and can quickly extract feature group information 4 from a large number of shape data groups 2.

[0046] First, the user inputs the shape data group 2, which is teaching data, and the corresponding shape code group 7 to the design device 1.

[0047] Next, the shape data group 2 and the corresponding shape code group 7 are temporarily stored in the encoder unit 3 as feature group information 4, which is information associating the shape data group 2 with the shape code group 7. Furthermore, when the user inputs a new shape code 8 to be generated to the design device 1, new shape data 6 corresponding to the new shape code 8 to be generated is constructed in the decoder unit 5 and output from the design device 1.

[0048] In the second embodiment, the decoder unit 5 inductively generates new shape data 6 different from the shape data group 2, which is the teacher data, based on the feature group information 4. In the first embodiment, the new shape code 8 to be generated is converted into a one-hot vector, but in the second embodiment, the input is a scalar value instead of a one-hot vector.

[0049] The scalar value may be any large or small value, but for ease of learning, a value normalized by the maximum value in the shape code set 7 (within the range of 0 to 1) is used as the input value.

[0050] In the first embodiment, feature group information 4 corresponding to a discrete shape code group 7 is stored, whereas in the second embodiment, feature group information 4 is stored in the decoder unit 5 in a continuous state such as a response surface based on a continuous scale, and new shape data 6 can be generated inductively from the feature group information 4.

[0051] Therefore, for example, when processing continuous data of n=1, 2, 3, 4, if one wishes to output n=1.5, by directly specifying 1.5 to the decoder unit 5, new shape data 6 similar to that in the first embodiment can be obtained.

[0052] FIG. 6 is a flowchart showing the process of the decoder unit 5 in the second embodiment.

[0053] In FIG. 6, the new shape code 8 is normalized (step S20), and the normalized scalar value is input to the design device 1 (step S21).

[0054] That is, in step S21, the decoder unit 5 associates the normalized scalar value with the feature group information 4. Then, the decoder unit 5 generates new shape data 6 from the feature group information 4 corresponding to the normalized scalar value (step S22) and outputs it (step S23).

[0055] In the second embodiment, the same effects as those in the first embodiment can be obtained.

[0056] A user can arbitrarily select either the first embodiment or the second embodiment depending on the difference in the design object.

[0057] In the above-mentioned first and second embodiments, the generated new shape data can be configured to be displayed.

[0058] 7 is a diagram showing a design device 1 having a display unit 9. New shape data 6 can be displayed on the display unit 9.

[0059] In the second embodiment, the new shape code 8 is normalized and the normalized scalar value is made to correspond to the feature group information 4 to generate the new shape data 6. However, it is also possible to convert the new shape code 8 using other formulas, not limited to normalization, and make it correspond to the feature group information 4 to generate the new shape data 6.

[0060] In other words, it is also possible to generate new shape data 6 using a shape data generation formula, typically a normalization formula.

[0061] Furthermore, in the design apparatus 1 and design method of the above-mentioned first and second embodiments, the feature group information 4 stored in the encoder unit 3 can be considered a database of existing shapes, and can be effectively utilized not only for new shape design but also for lineup design.

[0062] Furthermore, the present invention is not limited to the above-described first and second embodiments, but includes various modified examples.

[0063] For example, the above-mentioned first and second embodiments have been described in detail to clearly explain the present invention, and the present invention is not necessarily limited to having all of the configurations described.

[0064] In addition, it is possible to replace part of the configuration of one embodiment with the configuration of another embodiment, and it is also possible to add the configuration of another embodiment to the configuration of one embodiment.

[0065] Furthermore, it is possible to add, delete, or replace part of the configuration of each embodiment with other configurations. [Explanation of symbols]

[0066] 1 Design device, 2 Shape data group, 3 Encoder section, 4 Feature group information, 5 Decoder section, 6 New shape data, 7 Shape code group, 8 New shape code, 9 Display section

Claims

1. an encoder unit that receives shape data and a shape code that is a value corresponding to the shape data, calculates feature information of the shape data, and stores feature group information that is information that associates the feature information with the shape code; a decoder unit for generating new shape data corresponding to a new shape code from the feature group information; A design device comprising:

2. 2. The design apparatus according to claim 1, The design device is characterized in that the decoder unit synthesizes the feature group information corresponding to a plurality of the shape codes to generate the new shape data corresponding to the new shape code.

3. 2. The design apparatus according to claim 1, The decoder unit converts the new shape code using a shape data generation formula based on the shape code, and generates the new shape data corresponding to the new shape code converted from the feature group information.

4. 3. The design apparatus according to claim 2, The decoder unit converts the new shape code into a one-hot vector, modifies the one-hot vector in accordance with a synthesis ratio, and synthesizes the feature group information to generate the new shape data.

5. 4. The design apparatus according to claim 3, The encoder and decoder have a neural network structure using a convolutional layer, and the decoder normalizes the new shape code with a maximum value of the shape code using a normalization formula as the shape data generation formula, and generates the new shape data corresponding to the normalized new shape code.

6. 2. The design apparatus according to claim 1, A design device comprising a display unit for displaying the generated new shape data.

7. receiving shape data and a shape code which is a value corresponding to the shape data, calculating feature information of the shape data, and storing feature group information which is information in which the feature information and the shape code are associated with each other; A design method comprising the steps of: generating new shape data corresponding to a new shape code from said stored feature group information;

8. The design method according to claim 7, A design method comprising: synthesizing said feature group information corresponding to a plurality of said shape codes to generate said new shape data corresponding to said new shape code.

9. The design method according to claim 7, A design method comprising: converting the new shape code by a shape data generation formula based on the shape code; and generating the new shape data corresponding to the new shape code converted from the feature group information.

10. The design method according to claim 8, A design method comprising: converting the new shape code into a one-hot vector; modifying the one-hot vector according to a synthesis ratio; and synthesizing the feature group information to generate the new shape data.

11. 10. The design method according to claim 9, A design method comprising: normalizing the new shape code with a maximum value of the shape code using a normalization equation as the shape data generation equation; and generating the new shape data corresponding to the normalized new shape code.

12. The design method according to claim 7, A design method characterized in that the generated new shape data is displayed on a display unit.

Images