Drafting device

Abstract

This invention provides a drawing device that can create CAD drawings of the same shape using different CAD scripts. [Solution] By dividing the CAD script into a block section, a keyword section, and a parameter section and performing machine learning on each section, extracting shape features from the resulting data, and creating a surrogate model to extract predetermined shape features that target the same shape as the CAD data and are associated with those shape features, even if information defining a different CAD script from the aforementioned CAD script is input, it is possible to create a CAD drawing identical to the CAD data based on that information and the information of the associated shape features.

Description

Technical Field

[0004] , ,

[0005] , ,

[0001] The present invention relates to a drafting device. In particular, the present invention relates to an improvement for automatically creating CAD drawings. In this specification, the data of CAD drawings may sometimes be simply referred to as CAD data or drawing data.

Background Art

[0002] Conventionally, there is known a drafting device that uses a CAD script to draw a three-dimensional model (3D model). For example, in Patent Document 1, from the drawing data read from a drawing data file, attribute data of elements is extracted by executing a reading-side script, a data conversion module is activated, and a drafting script executable in a CAD system is created from the format of the read attribute data. By sequentially executing this drafting script, a CAD data file is obtained.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in conventional drafting devices, in order to create a CAD drawing having the same shape as the CAD drawing created by a CAD script, it is necessary to input the same CAD script, and it has not been possible to create the same CAD drawing with different CAD scripts. That is, even though it is possible to create the same CAD drawing by different procedures, it is necessary to input the same CAD script, and heretofore, a drafting device capable of creating the same CAD drawing by different procedures has not been constructed.

[0005] The present invention has been made in view of the above, and its objective is to provide a drawing device that can create CAD drawings of the same shape using different CAD scripts. [Means for solving the problem]

[0006] The present invention provides a solution for achieving the aforementioned objectives, which relates to a drawing device that creates CAD drawings based on a CAD script. This drawing device is characterized by including: a machine learning unit that divides a CAD script corresponding to CAD data into a block section that defines drawing instructions in the CAD script, a keyword section that specifies the shape of the figure specified by the block section, and a parameter section that defines the dimensions in the CAD data, and then performs machine learning on the division; a shape feature extraction unit that extracts shape features including the position coordinates in three-dimensional space of the CAD data from the data obtained by the machine learning; a machine learning model creation unit that creates a machine learning model for extracting predetermined shape features that target the same shape as the CAD data and are linked to the shape features; and a CAD drawing creation unit that, when information defining another CAD script different from the CAD script is input, creates a CAD drawing identical to the CAD data based on that information and the information of the linked shape features.

[0007] This specific requirement allows for the creation of CAD drawings that are consistent between the input information and the learned CAD scripts, making it possible to create CAD drawings of the same shape using different CAD scripts. [Effects of the Invention]

[0008] In this invention, a CAD script is divided into a block section, a keyword section, and a parameter section, and machine learning is performed on each section. Shape features are extracted from the resulting data, and a machine learning model is used to extract predetermined shape features associated with these shape features, thereby creating a CAD drawing identical to the CAD data. [Brief explanation of the drawing]

[0009] [Figure 1] Figure 1(a) is a block diagram showing a schematic of the drawing apparatus according to the embodiment, and Figure 1(b) is a diagram illustrating the process of extracting shape feature distributions based on CAD data. [Figure 2] This diagram illustrates the process of linking shape feature distributions extracted by a surrogate model when FE mesh data and CAE results are used as training data. [Figure 3] Figure 3(a) shows an example of CAD data and CAE results created when FE mesh data is input, and Figure 3(b) shows an example of CAD data, FE mesh data, and CAE results created when required performance data is input. [Modes for carrying out the invention]

[0010] Hereinafter, embodiments of the present invention will be described based on the drawings.

[0011] Figure 1(a) is a schematic block diagram of the drawing apparatus 1 according to this embodiment.

[0012] As shown in Figure 1(a), the drawing apparatus 1 according to this embodiment includes a CAD data acquisition unit 2, a tagging and classification unit 3, a machine learning unit (AI) 4, a shape feature extraction unit 5, a machine learning model creation unit 6, a CAD drawing creation unit 7, and a data output unit 8. However, the configuration of the drawing apparatus 1 is not limited to this.

[0013] The following explanation will describe each functional part with reference to Figures 1(b) and 2. Figure 1(b) is a diagram illustrating the process of extracting shape feature distributions based on CAD data (details will be described later). Figure 2 is a diagram illustrating the process of linking shape feature distributions extracted by a surrogate model (details will be described later) when FE mesh data and CAE results are used as training data.

[0014] The CAD data acquisition unit 2 is a functional unit that acquires CAD data (for example, CAD drawing data such as CATpart) representing the shape of an object for training a machine learning model (surrogate model, described later). The shape information acquired here is geometric information of each face, line, and point in three-dimensional space. Figure 1(b) shows the case where CAD data representing the shape of a machine part 10 having a disc part 11 and a cylindrical part 12 is acquired.

[0015] The tagging and classification unit 3 is a functional unit that classifies (automatically tags and classifies) a predetermined CAD script (e.g., CATIAVBS) into block parts, keyword parts, and parameter parts. As shown in Figure 1(b), the block part is the part that defines the drawing instructions in the CAD script (ASCII format information that becomes a command statement). The keyword part is the part that specifies the shape of the figure specified by the block part, defining the plane, lines, and points of the shape, and specifying the target of the block part's command. In particular, this keyword part is linked to the drawing order of the CAD drawing. The parameter part is the part that defines the dimensions in the CAD data (values ​​such as length, R, radius, etc. are entered).

[0016] The machine learning unit 4 is a functional unit that uses AI to learn CAD scripts, which are classified into block parts, keyword parts, and parameter parts by the tagging and classification unit 3, and the CAD data (shape data of the machine part 10) acquired by the CAD data acquisition unit 2. In this machine learning unit 4, a surrogate model (hereinafter referred to as the first surrogate model) is used to obtain the shape feature distribution shown in Figure 1(b) using the CAD data and CAD script as training data. For this reason, the tagging and classification unit 3 and the machine learning unit 4 correspond to the "machine learning unit (a machine learning unit in a broad sense, which divides a CAD script corresponding to CAD data into block parts that define drawing instructions in the CAD script, keyword parts that specify the shape of the figure specified by the block parts, and parameter parts that define the dimensions in the CAD data)" as defined in the present invention.

[0017] A surrogate model is generally a machine learning model that, when given multiple parameters defining the model to be simulated, outputs response values ​​related to the characteristics of the model to be simulated. This surrogate model can also include, for example, a neural network trained using machine learning. Furthermore, a surrogate model has an input layer into which multiple parameters defining the model to be simulated are input, an intermediate layer that extracts the shape features of the model to be simulated, and an output layer that outputs response values ​​related to the characteristics of the model to be simulated.

[0018] The shape feature extraction unit 5 is a functional unit that extracts shape features, including the position coordinates in the three-dimensional space of CAD data, from data obtained by machine learning in the machine learning unit 4. For example, it extracts shape features using latent variables. Specifically, it derives latent variables based on the content of the machine learning, and then extracts shape features based on these derived latent variables. The use of latent variables is a well-known method for extracting fundamental concepts by reducing the dimensionality of the data and aggregating observable variables into a model. In Figure 1(b), the distribution of three shape features (in the X, Y, and Z axes in Figure 1(b)) (hereinafter sometimes referred to as the shape feature distribution) is extracted. A brief explanation of shape features is given here. Shape features represent the characteristics and attributes of the data to be learned, and the shape feature extraction unit 5 extracts the necessary shape features by excluding unnecessary shape features from the dataset. Specific methods for extracting these shape features include known filtering methods, wrapping methods, and embedding methods.

[0019] The machine learning model creation unit 6 is a functional unit that creates a machine learning model (surrogate model) for extracting a predetermined shape feature distribution that is associated with the shape feature distribution.

[0020] Specifically, a surrogate model (hereinafter referred to as the second surrogate model) for extracting a shape feature amount distribution in the case where information defining a CAD script (for example, FE mesh data, CAE results, etc.) is input, with respect to a CAD drawing having the same shape as the CAD data acquired by the CAD data acquisition unit 2, is created (see FIG. 2). That is, the learning data (CAD data and CAD script) learned by the first surrogate model and the learning data (FE mesh data and CAE results) learned by the second surrogate model are learned by each surrogate model for the same shape. That is, the CAD data shape and the FE mesh shape use the same learning data group. In this case, in FIG. 2, a flange and a beam model are shown, but both are included in the learning data group. Thereby, the shape feature amount distribution obtained by the first surrogate model (the shape feature amount distribution located on the left side in FIG. 2) and the shape feature amount distribution obtained by the second surrogate model (the shape feature amount distribution located on the right side in FIG. 2) are associated, and it becomes possible to create the same CAD drawing even with different CAD scripts.

[0021] Specifically described, the CAD data and the script language are input to the first surrogate model, and the FE mesh data and the CAE results are input to the second surrogate model. As a result, the values of the shape feature amount distribution obtained by the first surrogate model (for example, a, b, c on each coordinate) and the values of the shape feature amount distribution obtained by the second surrogate model (for example, d, e, f on each coordinate) have vectors, and at this time, they are associated by coordinate transformation of (a, b, c) → (d, e, f). By this association, it becomes possible to create a CAD drawing having the same shape as the CAD data acquired by the CAD data acquisition unit 2 from the FE mesh and the components.

[0022] The CAD drawing creation unit 7 is a functional unit that newly constructs a CAD script by combining the surrogate models created in the machine learning model creation unit 6 and creates CAD data using this CAD script. As a result, when information different from the CAD script in the CAD data acquired by the CAD data acquisition unit 2 (for example, FE mesh data, CAE results, etc.) is input, based on this information and the information between the associated shape feature value distributions, a CAD drawing identical to the CAD data obtained by the machine learning can be created.

[0023] In this embodiment, as described above, the CAD script is divided into a block part, a keyword part, and a parameter part and machine-learned, and shape feature values (shape feature value distributions) are extracted from the data obtained thereby, and a machine learning model for extracting predetermined shape feature values associated with this shape feature value is used to create a CAD drawing identical to the CAD data. In other words, by creating a CAD drawing (CAD data) so that the input CAD script and the CAD script on the learned data are consistent, CAD drawings of the same shape can be created with different CAD scripts.

[0024] The data output unit 8 is a functional unit that outputs the CAD data (drawing data) created by the CAD drawing creation unit 7, and examples of the output destination of the CAD data include a monitor screen and a storage medium.

[0025] So far, a technology for converting from a mesh structure to CAD has not been constructed, but according to this embodiment, it is possible to construct this. Also, conventionally, a separate method of judgment was required for the shape output without relation to each performance (for example, yield strength, etc.), but according to this embodiment, this is no longer necessary.

[0026] Furthermore, conventionally, it was not possible to present a shape with respect to the required performance, but according to this embodiment, this becomes possible.

[0027] As specific examples, Figure 3(a) shows an example of CAD data and CAE results created when FE mesh data is input, and Figure 3(b) shows an example of CAD data, FE mesh data, and CAE results created when required performance data is input.

[0028] As shown in Figure 3(a), when FE mesh data is input to the drawing device 1, the aforementioned processing (processing using a surrogate model) will output CAD data (image data) and CAE results.

[0029] Furthermore, as shown in Figure 3(b), when the required performance data is input to the drawing device 1, the aforementioned processing outputs CAD data and CAE results similar to those shown in Figure 3(a), as well as FE mesh data similar to the input shown in Figure 3(a).

[0030] As described above, in this embodiment, the CAD script is divided into a block section, a keyword section, and a parameter section and subjected to machine learning. Shape features are extracted from the data obtained thereby, and a surrogate model is created to extract predetermined shape features that target the same shape as the CAD data and are associated with the shape features (shape feature distribution). As a result, even if information defining a different CAD script from the aforementioned CAD script is input, a CAD drawing identical to the CAD data can be created based on that information and the information of the associated shape features.

[0031] Furthermore, according to this embodiment, it is also possible to convert various types of information, such as CAD data, FE mesh data, and point cloud data, into CAD drawings by having the AI ​​learn them as a 3D structure and then verbalizing them.

[0032] -Other Embodiments- Furthermore, the present invention is not limited to the embodiments described above, and all modifications and applications are possible within the scope of the claims and equivalents thereof.

[0033] For example, algorithms such as convolutional neural networks (CNNs) and Gaussian processes are known to be used to create surrogate models, but the present invention is not limited to these.

[0034] Furthermore, in this invention, the parameters, hierarchy, and number of epochs when constructing the surrogate model are not particularly limited.

[0035] Furthermore, while CAD data, FE mesh data, and required performance data are effective inputs to the surrogate model, the present invention is not limited to these, and sketches (images), language, etc., may also be used as input data.

[0036] Furthermore, the present invention is not limited to industrial products, but can be applied to various things where performance and shape are related (for example, buildings, etc.).

[0037] Furthermore, in this invention, the required performance also includes graphs, numerical values, and language (for example, "soft" or "targeting Japanese women in their 20s").

[0038] Furthermore, the technical concept of the present invention also includes cases where the input and output are reversed in the embodiments described above. [Industrial applicability]

[0039] This invention is applicable to a drawing device that automatically creates CAD drawings. [Explanation of symbols]

[0040] 1. Drawing device 2. CAD data acquisition unit 3. Tagging and Classification Section 4. Machine Learning Department 5. Shape Feature Extraction Unit 6. Machine Learning Model Creation Department 7. CAD Drawing Creation Department 8. Data Output Section

Claims

[Claim 1] A drawing device that creates CAD drawings based on CAD scripts, A machine learning unit divides a CAD script corresponding to CAD data into a block section that defines drawing instructions in the CAD script, a keyword section that specifies the shape of the figure specified by the block section, and a parameter section that defines the dimensions in the CAD data, and then performs machine learning on these sections. A shape feature extraction unit extracts shape features, including the position coordinates in three-dimensional space of the CAD data, from the data obtained by the aforementioned machine learning. A machine learning model creation unit creates a machine learning model for extracting predetermined shape features that are linked to the aforementioned shape features and target the same shape as the CAD data, When information defining a CAD script different from the aforementioned CAD script is input, a CAD drawing creation unit creates a CAD drawing identical to the aforementioned CAD data based on that information and the information of the linked shape features. A drawing apparatus characterized by including the following:

Images