KR20200039050A - Method for providing digital drawing and digital drawing providing device - Google Patents

Abstract

The present invention relates a training data generation method and a training data generation system. The training data generation method for generating training data for an object of a plant drawing comprises: a shape data extraction step of receiving a plant drawing to extract shape data from an object on the plant drawing; a type data extraction step of extracting type data of the object based on the shape data of the object; and a training data generation step of generating training data of the object by merging the shape data and the type data.

Description

METHOD FOR PROVIDING DIGITAL DRAWING AND DIGITAL DRAWING PROVIDING DEVICE}

The present invention relates to a method for generating learning data and a system for generating learning data.

Due to the development of artificial intelligence, the field of use of deep learning technology is gradually expanding, and the type of data required varies depending on the purpose of use.

Deep learning technology in the field of plant engineering is used for the purpose of estimating plant basic specifications, diagnosing / predicting equipment failures, monitoring plant operation status, and analyzing plant operation data.

In order to successfully perform deep learning technology in the field of plant engineering, a large amount of training data that meets the objectives is required. However, manually building a large amount of training data takes a lot of time.

In addition, due to the occurrence of a human error, there may be a problem in accuracy, such as a mismatch with the basic design information of the plant model.

The present invention provides a learning data generation method and a learning data generation system capable of supporting a deep learning application tool to be used by a deep learning application by automatically extracting training data required for deep learning technology from a plant drawing and converting it into a format required for learning. The purpose.

On the other hand, the technical problems to be achieved in the present invention are not limited to the technical problems mentioned above, and other technical problems not mentioned are clearly understood by those skilled in the art from the following description. Will be understandable.

The learning data generation method according to an embodiment of the present invention is a learning data generation method for generating learning data for an object in a plant drawing, receiving the plant drawing and extracting shape data for extracting shape data from an object on the plant drawing step; A type data extraction step of extracting object type data based on the shape data of the object; And a learning data generation step of generating learning data of the object by merging the shape data and the type data. It includes.

The plant drawings may include 2D plant drawings and 3D plant drawings.

In the step of extracting the shape data, the object may include a symbol object and a text object, and the symbol object may include a device object and a pipe object.

In the step of extracting the shape data, the shape data may include size data and location data of the object on a plant drawing.

In the shape data extraction step, when the plant drawing is a 2D plant, a plurality of bounding boxes may be projected on the 2D plant drawing, and the shape of the object matching each bounding box may be extracted.

In the shape data extraction step, the size and position of the bounding box matched to the object on the plant drawing may be determined as size data and position data of the object.

In the shape data extraction step, a plurality of text objects matching the bounding box may be extracted from the plant drawing, and adjacent text objects among the plurality of text objects may be combined to extract shape data of the text object.

In the shape data extraction step, a straight line or a curve connecting a plurality of symbol objects on the plant drawing may be determined as the pipe object, and shape data of the pipe object may be extracted.

In the type data extraction step, shape data of the object may be matched with object data stored in a data library, and the type data of the matched object data may be extracted as the type data of the object.

In the step of generating the training data, when the plant drawing is a 2D plant, the training data is in an image format, and when the plant drawing is a 3D plant, the training data may be in an image format or a voxel format.

In addition, the learning data generation system according to an embodiment of the present invention is a learning data generation system for generating learning data for an object in a plant drawing, receiving the plant drawing and extracting shape data from an object on the plant drawing Data extraction unit; A type data extraction unit for extracting the type data of the object based on the shape data of the object; And a learning data generator configured to output learning data of the object by merging the shape data and the type data. It may include.

According to an embodiment of the present invention, it is possible to automatically support extraction of learning data required for deep learning technology from a plant drawing and converting it into a format required for learning, so that it can be used in a deep learning application tool.

On the other hand, the effects obtainable in the present invention are not limited to the above-mentioned effects, and other effects not mentioned will be clearly understood by those skilled in the art from the following description. Will be able to.

1 is an exemplary view showing the configuration of a learning data generation system according to an embodiment of the present invention,
2 is a flowchart illustrating a method for generating learning data according to an embodiment of the present invention,
3 is an exemplary view showing an example of extracting a symbol object in a shape data extraction step of a method for generating learning data according to an embodiment of the present invention in a 2D plant drawing,
FIG. 4 is an exemplary view showing learning data generated according to the example of FIG. 3 and a classification list thereof,
5 is an exemplary view showing another type of learning data generated according to the example of FIG. 3,
6 is an exemplary view showing an example of extracting a text object in a shape data extraction step of a method for generating learning data according to an embodiment of the present invention in a 2D plant drawing,
7 is an exemplary view showing learning data generated according to an example of FIG. 6,
8 is an exemplary view showing an example of extracting a wiring object in a shape data extraction step of a method for generating learning data according to an embodiment of the present invention in a 2D plant drawing;
9 is an exemplary view showing learning data generated according to an example of FIG. 8,
10 is an exemplary view showing a program for implementing a method for generating learning data according to an embodiment of the present invention in a 3D plant drawing,
11 is an exemplary view showing a process of generating 3D shape data in FIG. 10,
12 is an exemplary view showing 3D shape data extracted in FIG. 10,
13 is a program showing the image format learning data generation process in FIG. 10,
14 is an exemplary view showing three modes for selecting a viewpoint from 3D shape data,
15 is an exemplary view showing shape data of the image format learning data generated in FIG. 13;
16 is an exemplary view showing the final generated image format learning data,
17 is a program showing the process of generating voxel-type learning data in FIG. 10,
18 is an exemplary view showing shape data of the voxel type learning data generated in FIG. 17,
19 is an exemplary view showing the final generated image format learning data.

The embodiments of the present invention are provided to more fully describe the present invention to those of ordinary skill in the art, and the following embodiments can be modified in various other forms, and the scope of the present invention is as follows. It is not limited to the Examples. Rather, these examples are provided to make the present disclosure more faithful and complete, and to fully convey the spirit of the present invention to those skilled in the art.

In addition, in the following drawings, each configuration is exaggerated for convenience and clarity of description, and the same reference numerals in the drawings refer to the same elements. As used herein, the term “and / or” includes any and all combinations of one or more of the listed items.

The terminology used herein is used to describe a specific embodiment and is not intended to limit the present invention.

As used herein, singular forms may include plural forms unless the context clearly indicates otherwise. Also, as used herein, “comprise” and / or “comprising” specifies the shapes, numbers, steps, actions, elements, elements and / or the presence of these groups. And does not exclude the presence or addition of one or more other shapes, numbers, actions, elements, elements and / or groups.

Hereinafter, preferred embodiments of the present invention will be described in detail with reference to the accompanying drawings.

Hereinafter, a method of generating learning data according to an embodiment of the present invention will be described by classifying the embodiment according to a plant drawing.

The plant drawings may include 2D plant drawings and 3D plant drawings.

First, referring to FIGS. 1 to 9, a method for generating training data and a system for generating training data in a 2D plant drawing will be described.

1 is an exemplary view showing the configuration of a learning data generating system according to an embodiment of the present invention, FIG. 2 is a flowchart showing a learning data generating method according to an embodiment of the present invention, and FIG. 3 is the present invention in a 2D plant drawing An exemplary view showing an example of extracting a symbol object in a shape data extraction step of a method for generating learning data according to an embodiment of the present invention, and FIG. 4 is an exemplary view showing learning data generated according to an example of FIG. 3 and its classification list , Figure 5 is an exemplary view showing another type of learning data generated according to the example of Figure 3, Figure 6 is a text in the shape data extraction step of the learning data generation method according to an embodiment of the present invention in a 2D plant drawing 7 is an exemplary view showing an example of extracting an object, FIG. 7 is an exemplary view showing learning data generated according to an example of FIG. 6, and FIG. 8 is a 2D plant view. FIG. 9 is an exemplary view showing an example of extracting a wiring object in a shape data extraction step of a method for generating learning data according to an embodiment of the present invention, and FIG. 9 is an exemplary view showing learning data generated according to an example of FIG. 8.

First, referring to FIG. 1, the learning data generation system according to an embodiment of the present invention includes a shape data extraction unit 100, a type data extraction unit 200, a learning data generation unit 300, and a storage unit 400. It can contain.

In addition, referring to FIG. 2, a method for generating learning data according to an embodiment of the present invention may include a shape data extraction step (S10), a type data extraction step (S20), and a learning data generation step (S30).

3 and 4, the shape data extraction step (S10) may be performed by the shape data extraction unit 100 and project a bounding box for each object in a 2D plant drawing.

For example, in the step of extracting shape data (S10), a virtual bounding box (square or polygon, circle) is projected onto each symbol object, pipe object, and text object on a 2D plant drawing, and then the size of the virtual bounding box is adjusted. When all four sides of the virtual bounding box touch each object, the size of the virtual bounding box can be stopped and the shape data of the object in the area where the virtual bounding box is disposed can be extracted.

In addition, in the step of extracting the shape data (S10), it is preferable to define the coordinates of the center point of the virtual rectangle where the position and the size have been adjusted as the coordinates of the arrangement position of the recognized object.

Also, the size of the object can be defined through the corner coordinates (X, Y) of the bounding box.

That is, in the shape data extraction step (S10), the coordinate information of the symbol object, the pipe object, and the text object is clearly and quickly projected by projecting the virtual origin and the virtual square (or polygon, circle) on the input 2D plant drawing. It can be recognized separately.

In addition, referring to FIG. 6, in the shape data extraction step S10, a plurality of text objects may be extracted using a bounding box on a 2D plant drawing.

In addition, among the plurality of character objects, the arrangement positions are adjacent to each other, and text objects having the same array value (the same X value or the same Y value) are combined and grouped into a text object to extract shape data of the text object.

In addition, referring to FIG. 8, in the shape data extraction step (S10), after recognizing and recognizing a plurality of symbol objects and text objects in a 2D plant drawing, a geometric line connecting each symbol object is determined as a piping object, The shape data of the piping object can be extracted.

Thus, in the shape data extraction step (S10), the symbol object, the pipe object, and the text object can be clearly and quickly separated and extracted from the 2D plant drawing.

On the other hand, in the shape data extraction step (S10), a pre-processing step separates a work area (not shown), a title area (not shown), and an outline (not shown) on the 2D plant drawing, and excludes the title area and the outline, Geometry data for symbol objects, piping objects, and text objects can only be extracted from the work area.

Here, the work area is an area in which symbol objects, piping objects, and text objects to be practically recognized are arranged, and the heading area is an area in which text or tables, such as symbol objects, piping objects, and text object-related workers or construction information are shown. to be.

That is, in the shape data extraction step (S10) of the embodiment of the present invention, after removing an unnecessary area as a pre-processing step, an area to be analyzed can be specified, so that the extraction speed of the shape data can be improved.

On the other hand, although not shown, when the "Page-off connector" among the text objects is recognized on the 2D plant drawing in the shape data extraction step (S10), it is determined that other pages are connected, and other pages are linked to analyze the objects of other pages together can do.

Type data extraction step (S20) may be performed in the type data extraction unit 200, based on the shape data of the object extracted through the shape data extraction unit 100 to the storage unit (data library) 400 You can match the stored object information.

Although not illustrated, in the type data extraction step (S20), shape data of the object and shape data stored in the storage unit (data library) 400 may be compared to match the recognized object with the object information having the highest similarity.

Here, the type data may include code information based on a part number system stored in the storage unit (data library) 400.

Code information can be assigned a code to each data item for the object's equipment type, material, pressure rating, bonding method, etc., and multiple code combinations can be defined as a code list.

The actual value of a specific attribute can be found in the code list by referring to the generated code.

5, 7 and 9, the learning data generation step (S30) may be performed by the learning data generation unit 300, and the shape data stored in the storage unit 400 and the type data matching the data may be performed. By merging, it is possible to generate learning data for an object.

Meanwhile, the training data generated in the 2D plant drawing may be image format training data.

Here, the image format may be defined as a file such as txt, csv, xml, json, xml including image files such as jpg, png, and bmp.

In this way, it is possible to generate learning data for each deep learning.

Hereinafter, a method for generating training data and a system for generating training data in a 3D plant drawing will be described with reference to FIGS. 1 to 2 as well as FIGS. 10 to 19.

10 is an exemplary view showing a program implementing a method for generating learning data according to an embodiment of the present invention in a 3D plant drawing, FIG. 11 is an exemplary view showing a process of generating 3D shape data in FIG. 10, and FIG. 12 is Fig. 10 is an exemplary view showing 3D shape data extracted, Fig. 13 is a program showing an image format learning data generation process in Fig. 10, and Fig. 14 is an exemplary view showing three modes for selecting a viewpoint from 3D shape data , FIG. 15 is an exemplary diagram showing shape data of the image format training data generated in FIG. 13, FIG. 16 is an example diagram showing the final generated image format training data, and FIG. 17 is generating voxel format training data in FIG. 10 A program showing the process, FIG. 18 is an exemplary view showing shape data of the voxel type learning data generated in FIG. 17, and FIG. 19 is the final generation It is an example showing the image format learning data.

In the present invention, the plant 3D shape learning data generation system for deep learning may include a shape data extraction module, a voxel format learning data generation module, and an image format learning data generation module.

First, by utilizing the 'shape data extraction module', the fitting shape of AutoCAD Plant 3D can be extracted into the shape of the STL structure and converted into the shape of the OBJ structure.

Thereafter, by using the extracted OBJ shape, each of the deep learning CAD data can be generated through the 'image format learning data generation module' and the 'voxel type learning data generation module'.

Referring to FIG. 10, the shape data extraction module may be configured as a plugin-type module of AutoCAD Plant 3D for extraction of shape (OBJ) and non-shape (XML) files of a 3D CAD model.

In the shape data extraction step (S10) and the type data extraction step (S20), a 3D CAD model may be extracted using a shape data extraction module.

Plant 3D shape data of AutoCAD Plant 3D, which contains basic design information of a 3D plant model, can be extracted as a file of an XML structure that contains mesh files of OBJ structure, shape information, and type information.

The input / output data of the module is as follows.

-Input data: specification name, shape extraction path

-Output data: OBJ structure shape data, XML structure non-shape data

Here, since the AutoCAD Plant 3D system does not support the output of the OBJ, it is preferable to output the shape of the STL structure and convert it to the OBJ structure.

Meanwhile, referring to FIG. 11, the shape data extraction module may be provided to a user as a program.

After loading the module DLL file through the command input (NETLOAD) in AutoCAD Plant 3D in the corresponding program, the module can be executed through the command input (CMD_OBJEXTRACTOR).

In addition, you can select one of the specifications supported by the AutoCAD Plant 3D system by using the 'Input Spec Information' button, and input the path of the output shape using the 'Input Object Output Path' button, and the 'Output Object Output' button You can output OBJ file and XML file by using.

The extracted 3D shape data OBJ is as shown in FIG. 12.

On the other hand, as described above, in the learning data generation step (S30) of the present invention, it is possible to generate learning data in an image format or voxel format in a 3D plant drawing.

13 to 16, in the learning data generation step (S30) of the learning data generation method of the embodiment of the present invention, learning data in an image format may be generated.

The program UI shown in FIG. 13 is a standalone-type module for the purpose of generating learning data of depth images and RGB images using a 3D CAD model extracted with a learning data generation module of an image format. .

In the learning data generation step (S30), the learning data in the image format may be generated using the 'image data training module'.

The input / output data of the module is as follows.

-Input data: OBJ structure shape data, learning data extraction path

-Output data: Image format (depth, RGB) learning data (* .txt)

Here, scenes at various viewpoints of the three-dimensional shape of the OBJ structure are converted into images (depth, RGB), and output as a text structure file that is easy to use for tools such as deep learning.

Meanwhile, the following functions are provided for the accuracy of the resulting image.

-Image height and width size setting function

-Setting function of image noise generation

-3D model viewpoint selection function

As shown in FIG. 14, in the image format learning data generation module, three modes (Sphere, Surround, Dodeca) may be supported for selection of viewpoints.

The Sphere mode can set a viewpoint at the vertex of a regular polyhedron (diversity of viewpoints according to the level), and the Surround mode can set the altitude of the viewpoint and the number of divisions in the azimuth direction and the maximum minimum angle of the viewpoint in the elevation direction.

In addition, in the Dodeca mode, it is possible to set a point of view at the dodecahedron vertex.

Through the above-described process, shape data in an image format (depth, RGB) as shown in FIG. 15 may be extracted, and training data in an image format as shown in FIG. 16 may be generated.

Also, referring to FIGS. 17 to 19, in the learning data generation step (S30) of the learning data generation method of the embodiment of the present invention, voxel-type learning data may be generated.

In the process of generating voxel type learning data, a command prompt (cmd.exe) may be executed as shown in FIG. 17 to input commands according to the following rules.

[Voxel type learning data extraction module] [OBJ file] [Output file]

In the learning data generation step (S30), a 'voxel-type learning data generation module' may be used to generate voxel-type learning data.

The input / output data of the module is as follows.

-Input data: OBJ structure shape data, learning data extraction path

-Output data: voxel type learning data (* .txt)

Here, the VTK open source can be used to convert the three-dimensional shape of the OBJ structure into a voxel model and output it as a text structure file that is easy to use for tools such as deep learning.

That is, according to the present invention, in order to apply intelligent information technology such as deep learning to engineering application fields such as design automation, it is possible to convert 3D CAD data, which is a main storage medium of 3D design information, into a form required for deep learning learning, and 3D By integrating CAD system application SW development technology, 3D CAD data conversion technology, and deep learning related technology, it is possible to provide consistent procedures and functions for extracting the data necessary for learning from the plant 3D CAD system.

Through the above-described process, shape data in the form of voxels as shown in FIG. 18 may be extracted, and learning data in the form of images as shown in FIG. 19 may be generated.

What has been described above is only one embodiment for implementing the method for generating learning data according to the present invention, and the present invention is not limited to one embodiment, and the subject matter of the present invention is claimed in the following claims. Anyone who has ordinary knowledge in the field to which the present invention pertains without departing will have the technical spirit of the present invention to the extent that various changes can be implemented.

100: extraction unit 200: type data extraction unit
300: learning data generation unit 400: storage unit

Claims (12)

In the learning data generation method for generating learning data for an object of the plant drawing,
A shape data extraction step of receiving the plant drawing and extracting shape data from objects on the plant drawing;
A type data extraction step of extracting object type data based on the shape data of the object; And
A learning data generation step of generating learning data of the object by merging the shape data and the type data; Method for generating learning data comprising a.
According to claim 1,
The plant drawing includes a 2D plant drawing and a 3D plant drawing.
According to claim 2,
In the step of extracting the shape data
The object includes a symbol object and a text object,
The symbol object is a method for generating learning data including a device object and a piping object.
According to claim 3,
In the step of extracting the shape data
The shape data is a method of generating learning data including size data and position data of the object on a plant drawing.
According to claim 4,
In the step of extracting the shape data
If the plant drawing is a 2D plant,
Projecting a plurality of bounding boxes on the 2D plant drawing,
Method for generating training data to extract the shape of an object matching each bounding box.
The method of claim 5,
In the step of extracting the shape data
A method of generating learning data for determining the size and position of the bounding box matched to the object as the size data and position data of the object on the plant drawing.
The method of claim 5,
In the step of extracting the shape data
Extracting a plurality of text objects matching the bounding box on the plant drawing,
A method for generating learning data by extracting shape data of the text object by combining adjacent text objects among the plurality of text objects.
The method of claim 7,
In the step of extracting the shape data
The straight line or curve connecting a plurality of symbol objects on the plant drawing is determined as the piping object,
Method for generating learning data to extract shape data of the piping object.
According to claim 2,
In the type data extraction step
Match the shape data of the object to the object data stored in the data library,
A method of generating learning data that extracts the type data of the matched object data as the type data of the object.
The method of claim 9,
In the step of generating the learning data
When the plant drawing is a 2D plant, the learning data is in an image format,
When the plant drawing is a 3D plant, the learning data is an image format or a voxel format.
In a computer-readable recording medium,
A recording medium in which a program for implementing the method for generating learning data according to any one of claims 1 to 10 is recorded.
In the learning data generation system for generating learning data for the object of the plant drawing,
A shape data extracting unit that receives the plant drawing and extracts shape data from objects on the plant drawing;
A type data extraction unit for extracting the type data of the object based on the shape data of the object; And
A learning data generation unit that merges the shape data and the type data to output learning data of the object; Learning data generation system comprising a.

Images