KR20240055208A - Method, Apparatus and Computer Program for Optimal Generative Design based on Reference Data and Implicit Neural Representation - Google Patents

Abstract

An embodiment of the present invention relates to a generative design method performed by a computing device, comprising the process of applying generative design to reference data to obtain a plurality of new designs having similarity to the reference data. A design method can be provided.

Description

{Method, Apparatus and Computer Program for Optimal Generative Design based on Reference Data and Implicit Neural Representation}

The present invention relates to an optimized generative design method, device, and computer program based on reference data and implicit neural representations.

In various industries at home and abroad, we are trying to improve product competitiveness by automating the product development process using artificial intelligence (AI) to shorten product development time and minimize costs.

The use of synthetic data is essential to learn artificial intelligence models in data-scarce environments. Synthetic data refers to data artificially created through various automated technologies such as simulation.

However, in the case of 3D engineering data, it is necessary to generate physically meaningful synthetic data while understanding the engineering domain, which requires more technical know-how than other data, so a large amount of 3D data can be used for artificial intelligence learning in actual product development sites. We are having difficulty constructing engineering data.

Registered Patent Publication No. 10-2079027 (Topology optimal design method using deep learning technology)

Embodiments of the present invention provide an optimized generative design method, device, and computer program based on reference data and implicit neural expressions that can generate various designs for products that require small quantity production of a large variety and products that require consideration of aesthetics. It has a purpose.

An embodiment of the present invention is an optimized generative design based on reference data and implicit neural representation that can automatically generate multiple new designs for a product even when there is a small amount of reference data or no reference data for the product. The purpose is to provide methods, devices, and computer programs.

Embodiments of the present invention provide an optimized generative design method, device, and computer program based on reference data and implicit neural expressions that can provide new designs for 3D products using images or modeling of 3D products. The purpose is to provide

An embodiment of the present invention maps all seed data into a low-dimensional latent space through deep learning and then processes the mapped data within the latent space to generate synthetic data for learning an artificial intelligence model. The purpose is to provide optimized generative design methods, devices, and computer programs based on reference data and implicit neural representations.

Embodiments of the present invention provide an optimized generative design method, device, and computer program based on reference data and implicit neural representation that can generate and explore design data in real time using design mapping data in a low-dimensional space. It has a purpose.

One embodiment of the present invention is an optimized generative design method, device, and device based on reference data and implicit neural representation, which generates a new design that is 2D/3D composite data for developing a product or virtual product using reference data. The purpose is to provide computer programs.

One embodiment of the present invention is an optimized generative design method and device based on reference data and implicit neural representation that can explore various designs that satisfy design requirements in a low-dimensional space (z) called design space. The purpose is to provide and computer programs.

The purpose of one embodiment of the present invention is to provide an optimized generative design method, device, and computer program based on reference data and implicit neural representation that can generate a design that satisfies the engineering requirements of a product.

One embodiment of the present invention provides an optimized generative design method, device, and computer program based on reference data and implicit neural expression, which provides a method for auto-labeling the generated new design through automation of CAE analysis. It has a purpose.

One embodiment of the present invention generates a new design and provides design technology for products using the new design using an artificial intelligence model learned using the generated new design (e.g., 3D synthetic data), reference data and The purpose is to provide optimized generative design methods, devices, and computer programs based on implicit neural representations.

One embodiment of the present invention is a reference data that provides a Deep Generative Design service that can create a new design using a small amount of reference data and train the created new design in an artificial intelligence model. The purpose is to provide optimized generative design methods, devices, and computer programs based on implicit neural representations.

An embodiment of the present invention is an optimized generative design method and device based on reference data and implicit neural expressions that can simultaneously solve the problem of lack of data and lack of AI experts occurring in product development sites in the manufacturing industry. The purpose is to provide and computer programs.

One embodiment of the present invention is to effectively explore new design ideas that have not been considered in the past at the concept design stage and at the same time propose an optimized design considering various performance at a fast speed close to real-time, using reference data and implicit The purpose is to provide optimized generative design methods, devices, and computer programs based on neural representation.

An embodiment of the present invention relates to a generative design method performed by a computing device, comprising the process of applying generative design to reference data to obtain a plurality of new designs having similarity to the reference data. A design method can be provided.

An embodiment of the present invention includes a process of filtering the plurality of new designs; A process of obtaining a plurality of additional new designs generated using a design generation model based on the plurality of filtered new designs; a process of filtering the plurality of additional new designs; and a step of obtaining the plurality of new designs by applying the generative design using the filtered plurality of additional new designs as the plurality of reference data.

An embodiment of the present invention further includes a process of acquiring the reference data, wherein the first reference data includes 2D reference data or 3D reference data, and the 2D reference data includes shape image data, pattern data, and skeleton data. , graph data, SDF data (distance information), depth data, implicit function data, grammar, and multi-view data obtained from each type of data, and the 3D reference data includes CAD data, voxel data, mesh data, A generative design method can be provided that includes at least one of point cloud data, Octree data, shape parameter data, pattern data, skeleton data, graph data, SDF data (distance information), and implicit functions (functional expression or deep learning model). there is.

In an embodiment of the present invention, the generative design includes (i) a topology optimization methodology using at least one of SIMP, ESO, BESO, LevelSet, MMC, and deep learning, (ii) a dimension and shape optimization methodology, and (iii) ) A generative design method using at least one of the parametric design methodologies may be provided.

An embodiment of the present invention is a generative design method in which the objective function of the generative design is composed of a performance function portion corresponding to the requirements input by the user and a similarity function portion that determines similarity to the reference data. can be provided.

An embodiment of the present invention is a generative design in which the sensitivity function of the generative design is composed of a part that differentiates the performance function corresponding to the requirements entered by the user and a similarity function part that determines similarity with the reference data. Design methods can be provided.

In an embodiment of the present invention, the process of filtering the plurality of new designs or additional new designs includes at least one of a performance filtering process and a similarity filtering process of the new design, and the performance filtering process is performed according to requirements set by the user. Filters low-performance new designs through evaluation, and the similarity filtering process evaluates and filters the similarity on a high-dimensional domain where the shape of the new design exists, or maps the shape of the new design onto a low-dimensional domain to determine the similarity. It can provide a generative design method that evaluates and filters.

Embodiments of the present invention may provide a generative design method in which the design generation model includes at least one of a deep learning model, a Boolean model, a morphing model, and an interpolation model.

An embodiment of the present invention uses the plurality of new designs to learn an implicit neural expression model, and maps the plurality of new designs into a low-dimensional, continuous, parametric space (z) to generate a plurality of design mapping data. The process of acquiring; And using the plurality of design mapping data in the low-dimensional, continuous, parametric space (z), at least one of a design exploration process, a design interpolation process, a design performance prediction process, a design optimization process, and a design reverse engineering process. A generative design method including:

An embodiment of the present invention is a generative design where, in the design optimization process, topology optimization is performed by applying a topology optimization technique to the plurality of design mapping data in the low-dimensional, continuous, parametric space (z). A method can be provided.

Embodiments of the present invention include a processor; network interface; Memory; and a computer program loaded into the memory and executed by the processor, wherein the computer program applies generative design to reference data for the product to create a plurality of new products having similarity to the reference data. A generative design device may be provided, including instructions for obtaining a design.

Embodiments of the present invention may be combined with a computing device to execute a process of applying generative design to reference data for a product to obtain a plurality of new designs having similarities to the reference data, a computer-readable record. A generative design computer program stored on a medium may be provided.

An embodiment of the present invention is a method performed by a computing device, wherein a plurality of designs (X) for a product are mapped to a low-dimensional latent space (z) using an implicit neural expression model to generate a plurality of design mapping data. The process of obtaining (Z); and a generative design method using the plurality of design mapping data in the latent space (z), including at least one of a design exploration process, a design interpolation process, a design performance prediction process, a design optimization process, and a design reverse engineering process. can be provided.

In an embodiment of the present invention, the design search process searches for first design mapping data (Z ₁ ) among the plurality of design mapping data (Z) in the latent space (z), and searches for the first design mapping data (Z 1 ) in the latent space (z). A generative design method may be provided to obtain a first design (X ₁ ) corresponding to Z ₁ ).

In an embodiment of the present invention, the design interpolation process selects N data from the plurality of design mapping data (Z) in the latent space (z) and interpolates them into second design mapping data (Z ₂ ) through an interpolation method. It is possible to provide a generative design method for obtaining a second design (X ₂ ) corresponding to the interpolated second design mapping data (Z ₂ ).

In an embodiment of the present invention, the design prediction process includes the plurality of design mapping data (Z) of the latent space (z) and the plurality of designs (X) corresponding to the plurality of design mapping data (Z). A generative design that uses a performance prediction model learned with performance data (Y) to predict predicted performance data (Y ₃ ') through third design mapping data (Z ₃ ) to a third design (X ₃ ). A method can be provided.

In an embodiment of the present invention, the design optimization process is based on the performance data (Y) of the plurality of designs (X), and the fourth design (X _{4 ) corresponding to the fourth design (X 4} ) within the latent space (z). A generative design method can be provided that optimizes design mapping data (Z ₄ ) and obtains an optimized fourth design (X ₄ ').

In an embodiment of the present invention, the reverse design process is a reverse design model learned with performance data (Y) of the latent space (z) and the plurality of design mapping data (Z) corresponding to the performance data (Y). Using , obtain the fifth design mapping data (Z ₅ ) corresponding to the fifth performance data (Y ₅ ), and obtain the fifth design data (X ₅ ) corresponding to the fifth design mapping data (Z ₅ ). A generative design method can be provided.

Embodiments of the present invention provide an optimized generative design method, device, and computer program based on reference data and implicit neural representation, which provides a new design with external aesthetics while ensuring engineering performance regardless of the type of reference data. can be provided.

An embodiment of the present invention uses reference data and implicit neural representation to create a new 2-dimensional or 3-dimensional design for a product using not only 2-dimensional or 2.5-dimensional data about the product, but also 3-dimensional data about the product as reference data. Based on optimized generative design methods, devices, and computer programs can be provided.

Embodiments of the present invention are an optimized generative design method, device, and device based on reference data and implicit neural representation that can derive a new three-dimensional design for a product using easily collectable two-dimensional data on the product. Computer programs can be provided.

Embodiments of the present invention are an optimized generative design method, device, and computer program based on reference data and implicit neural expressions that enable new designs by reflecting reference data on multi-views of the product when designing a concept for a product. can be provided.

An embodiment of the present invention provides an optimized generative design method based on reference data and implicit neural expressions that provides a new three-dimensional design of a product that satisfies product performance and has aesthetics for multi-view desired by users; Devices and computer programs can be provided.

An embodiment of the present invention maps all seed data into a low-dimensional latent space through deep learning and then processes the mapped data within the latent space to generate synthetic data for learning an artificial intelligence model. Optimized generative design methods, devices, and computer programs based on reference data and implicit neural representations can be provided.

Embodiments of the present invention provide an optimized generative design method, device, and computer program based on reference data and implicit neural representation that can generate and explore design data in real time using design mapping data in a low-dimensional space. can do.

One embodiment of the present invention is an optimized generative design method, device, and device based on reference data and implicit neural representation, which generates a new design that is 2D/3D composite data for developing a product or virtual product using reference data. Computer programs can be provided.

One embodiment of the present invention is an optimized generative design method and device based on reference data and implicit neural representation that can explore various designs that satisfy design requirements in a low-dimensional space (z) called design space. and computer programs can be provided.

One embodiment of the present invention can provide an optimized generative design method, device, and computer program based on reference data and implicit neural representation that can generate a design that satisfies the engineering requirements of a product.

One embodiment of the present invention provides an optimized generative design method, device, and computer program based on reference data and implicit neural expression, which provides a method for auto-labeling the generated new design through automation of CAE analysis. can do.

One embodiment of the present invention generates a new design and provides design technology for products using the new design using an artificial intelligence model learned using the generated new design (e.g., 3D synthetic data), reference data and Optimized generative design methods, devices, and computer programs based on implicit neural representations can be provided.

One embodiment of the present invention is a reference data that provides a Deep Generative Design service that can create a new design using a small amount of reference data and train the created new design in an artificial intelligence model. and can provide optimized generative design methods, devices, and computer programs based on implicit neural representations.

One embodiment of the present invention is an optimized generative design method and device based on reference data and implicit neural expressions that can simultaneously solve the problem of lack of data and lack of AI experts occurring in product development sites in the manufacturing industry. and computer programs can be provided.

One embodiment of the present invention is to effectively explore new design ideas that have not been considered in the past at the concept design stage and at the same time propose an optimized design considering various performance at a fast speed close to real-time, using reference data and implicit Optimized generative design methods, devices, and computer programs based on neural representation can be provided.

Figure 1 is a hardware configuration diagram of an optimized generative design device based on reference data and implicit neural representation according to an embodiment of the present invention.
Figure 2 is a flowchart of an optimized generative design method based on reference data and implicit neural representations according to an embodiment of the present invention.
Figure 3 is a block diagram of an optimized generative design method based on reference data and implicit neural representation according to an embodiment of the present invention.
Figure 4 is a conceptual diagram of an optimized generative design method based on reference data and implicit neural representation according to an embodiment of the present invention.
Figure 5 is a flowchart of a design creation method according to an embodiment of the present invention.
Figure 6 is a flowchart of a process for obtaining reference data in an optimized generative design method based on reference data and implicit neural representations according to an embodiment of the present invention.
Figures 7(a) to 7(d) show examples of preprocessed images generated by preprocessing input data.
Figure 8 shows various examples of new designs created using pattern images as reference data.
Figure 9 shows various examples of new designs created using a skeleton image as reference data.
Figure 10 shows various examples of new designs created using a view image or cross-sectional image as reference data.
Figure 11 shows an example of design interpolation in latent space.

Since various changes can be made to the embodiments of the present invention, the embodiments will be illustrated in the drawings and described in detail in the specification. However, this is not intended to limit the embodiments of the present invention to specific forms, and includes all changes, equivalents, or substitutes included in the spirit and technical scope of the present invention.

The terminology used herein is for describing embodiments and is not intended to limit the invention. As used herein, singular forms also include plural forms, unless specifically stated otherwise in the context. As used in the specification, “comprises” and/or “comprising” does not exclude the presence or addition of one or more other elements in addition to the mentioned elements. Like reference numerals refer to like elements throughout the specification, and “and/or” includes each and every combination of one or more of the referenced elements. Although “first”, “second”, etc. are used to describe various components, these components are of course not limited by these terms. These terms are merely used to distinguish one component from another. Therefore, it goes without saying that the first component mentioned below may also be a second component within the technical spirit of the present invention. In this specification, the meaning of including at least one of A, B, and C includes all various combinations of one or two or more of A, B, and C. For example, {A}, {B}, {C}, {A, B}, {A, C}, {B, C}, {A, B, C}.

Unless otherwise defined, all terms (including technical and scientific terms) used in this specification may be used with meanings commonly understood by those skilled in the art to which the present invention pertains. Additionally, terms defined in commonly used dictionaries are not to be interpreted ideally or excessively unless clearly specifically defined.

As used in the specification, the term “unit” or “module” refers to a hardware component such as software, FPGA, or ASIC, and the “unit” or “module” performs certain roles. However, “part” or “module” is not limited to software or hardware. A “unit” or “module” may be configured to reside on an addressable storage medium and may be configured to run on one or more processors. Thus, as an example, a “part” or “module” refers to components such as software components, object-oriented software components, class components, and task components, processes, functions, properties, Includes procedures, subroutines, segments of program code, drivers, firmware, microcode, circuits, data, databases, data structures, tables, arrays, and variables. The functionality provided within components and “parts” or “modules” can be combined into smaller components and “parts” or “modules” or into additional components and “parts” or “modules”. Could be further separated.

Spatially relative terms such as “below”, “beneath”, “lower”, “above”, “upper”, etc. are used as a single term as shown in the drawing. It can be used to easily describe the correlation between a component and other components. Spatially relative terms should be understood as terms that include different directions of components during use or operation in addition to the directions shown in the drawings. For example, if a component shown in a drawing is flipped over, a component described as “below” or “beneath” another component will be placed “above” the other component. You can. Accordingly, the illustrative term “down” may include both downward and upward directions. Components can also be oriented in other directions, so spatially relative terms can be interpreted according to orientation.

Hereinafter, an optimized generative design method, device, and computer program based on reference data and implicit neural representation according to an embodiment of the present invention will be described in detail with reference to the attached drawings.

Figure 1 is a hardware configuration diagram of an optimized generative design device based on reference data and implicit neural representation according to an embodiment of the present invention.

Referring to FIG. 1, an optimized generative design device 100 (hereinafter referred to as “computing device 100”) based on reference data and implicit neural representation according to an embodiment of the present invention includes one or more processors 110. , it may include a memory 120 that loads the computer program 141 executed by the processor 110, a communication interface 130, and a storage 140 that stores the computer program 141.

Here, the optimized generative design device 100 based on reference data and implicit neural representation according to an embodiment of the present invention is not limited to the components shown in FIG. 1 and includes other general-purpose components. More may be included.

In various embodiments, the computing device 100 refers to all types of hardware devices including at least one processor 110, and depending on the embodiment, it is understood to also encompass software configurations that operate on the hardware device. It can be.

The computing device 100 may be understood as including, but is not limited to, servers, smartphones, tablet PCs, desktops, laptops, and user clients and applications running on each device.

Each process of the optimized generative design method based on reference data and implicit neural representation according to an embodiment of the present invention is described as being performed by the computing device 100, but the subject of each process is not limited thereto. Depending on the embodiment, at least part of each process may be performed on different computing devices.

The processor 110 controls the overall operation of each component of the computing device 100. The processor 110 includes a Central Processing Unit (CPU), Micro Processor Unit (MPU), Micro Controller Unit (MCU), Graphic Processing Unit (GPU), or any other type of processor well known in the art of the present invention. It can be.

In addition, the processor 110 may perform operations on at least one application or program to execute an optimized generative design method based on reference data and implicit neural representation according to an embodiment of the present invention, and is a computing device. 100 may have one or more processors.

In addition, the processor 110 includes RAM (Random Access Memory, not shown) and ROM (Read-Only Memory, not shown) that temporarily and/or permanently store internally processed signals (or data). It may further include. Additionally, the processor 110 may be implemented in the form of a system on chip (SoC) that includes at least one of a graphics processing unit, RAM, and ROM.

Memory 120 stores various data, commands and/or information. The memory 120 may load the computer program 141 from the storage 140 to execute an optimized generative design method based on reference data and implicit neural representations according to an embodiment of the present invention. When the computer program 141 is loaded into the memory 120, the processor 110 can perform the method by executing one or more instructions constituting the computer program 141. The memory 120 may be implemented as a volatile memory such as RAM, but the technical scope of the present disclosure is not limited thereto.

The bus (BUS) provides communication functions between components of the computing device 100. A BUS can be implemented as various types of buses, such as an address bus, data bus, and control bus.

The communication interface 130 supports wired and wireless Internet communication of the computing device 100. Additionally, the communication interface 130 may support various communication methods other than Internet communication. For example, the communication interface 130 may support at least one of short-range communication, mobile communication, and broadcast communication methods. To this end, the communication interface 130 may be configured to include a communication module well known in the technical field of the present invention. In some embodiments, communication interface 130 may be omitted.

Storage 140 may store the computer program 141 non-temporarily. The storage 140 is a non-volatile memory such as Read Only Memory (ROM), Erasable Programmable ROM (EPROM), Electrically Erasable Programmable ROM (EEPROM), flash memory, a hard disk, a removable disk, or a device well known in the technical field to which the present invention pertains. It may be configured to include any known type of computer-readable recording medium.

The computer program 141, when loaded into the memory 120, includes one or more instructions that cause the processor 110 to perform an optimized generative design method based on reference data and implicit neural representations according to an embodiment of the present invention. It can be included. That is, the processor 110 can perform the method according to various embodiments of the present invention by executing the one or more instructions.

The processes of the optimized generative design method based on reference data and implicit neural representation according to embodiments of the present invention may be implemented directly in hardware, implemented as a software module executed by hardware, or implemented by a combination thereof. You can. The software module may be RAM (Random Access Memory), ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), Flash Memory, hard disk, removable disk, CD-ROM, or It may reside on any type of computer-readable recording medium well known in the art to which the present invention pertains.

The processes of the optimized generative design method based on reference data and implicit neural representation according to an embodiment of the present invention may be implemented as a program (or application) and stored in a medium to be executed in combination with a computer, which is hardware.

Components of the invention may be implemented as software programming or software elements, and similarly, embodiments may include various algorithms implemented as combinations of data structures, processes, routines or other programming constructs, such as C, C++, , may be implemented in a programming or scripting language such as Java, assembler, etc. Functional aspects may be implemented as algorithms running on one or more processors.

Below, an optimized generative design method based on reference data and implicit neural representation according to an embodiment of the present invention will be described.

Figure 2 is a flowchart of an optimized generative design method based on reference data and implicit neural representations according to an embodiment of the present invention. Figure 3 is a block diagram of an optimized generative design method based on reference data and implicit neural representation according to an embodiment of the present invention. Figure 4 is a conceptual diagram of an optimized generative design method based on reference data and implicit neural representation according to an embodiment of the present invention.

Referring to FIG. 2, the optimized generative design method based on reference data and implicit neural representation according to an embodiment of the present invention includes a design creation process (S10), a design exploration process (S20), and design prediction. It may consist of at least one or a combination of two or more of the process (S30) and the design optimization process (S40).

Referring to FIG. 3, the generator (P10) can perform the design creation process (S10), the searcher (P20) can perform the design search process (S20), and the predictor (P30) can perform the design search process (S20). The prediction process (S30) can be performed, and the optimizer (P40) can perform the design optimization process (S40).

At least one of the generator (P10), searcher (P20), predictor (P30), and optimizer (P40) according to an embodiment of the present invention is directly implemented in hardware, is implemented as a software module executed by hardware, or It can be implemented by combining these

In the S10 process, the creator (P10) can create a large amount of new designs using generative design technology based on a small amount of reference data. In this case, the generator (P10) can create a new design that resembles the user's reference data (or past design data) and is engineering-excellent.

In the S20 process, the searcher (P20) can learn an artificial intelligence model that maps a high-dimensional new design to a low-dimensional latent space. Artificial intelligence models may include deep learning models. Multiple new designs can be mapped into multiple design mapping data in a low-dimensional latent space using a deep learning model. Design mapping data can have continuous values in a low-dimensional latent space.

All points in the latent space corresponding to multiple design mapping data can be restored into multiple different designs (e.g., 3D design plans). Therefore, the searcher (P20) can search for high-dimensional designs by searching design mapping data with continuous values in the latent space, and can cluster similar designs based on design mapping data located nearby in the latent space. there is. In this case, the discovered design can be expressed as an exploration design.

The explorer (P20) can create new designs needed by the user in real time through design of experiments (DOE) required for CAE analysis in the latent space, and find new designs that are more efficiently optimized in the latent space.

In the S30 process, the predictor (P30) can receive CAE analysis data or experimental data as input and learn a performance prediction model that predicts performance in the latent space. In this case, the performance prediction model is an artificial intelligence model and may include a deep learning model. Alternatively, the predictor (P30) can learn a performance prediction model that predicts performance in the latent space by receiving new designs and experimental design data generated through design of experiments (DOE) as input.

In this case, the predictor (P30) can predict the performance of a new design and/or all design mapping data that can be explored in the latent space in real time using the learned performance prediction model.

In the S40 process, the optimizer (P40) can learn an artificial intelligence model that finds or creates a new optimized design in the latent space in real time when the user inputs the desired requirements (e.g., target performance). Artificial intelligence models may include deep learning models.

The optimizer (P40) can create a new optimized design in real time using a learned artificial intelligence model, and further optimizes it by fine-tuning the corresponding optimal design mapping data around the potential space where the proposed optimal new design is located. You can create a new design.

The optimized generative design method based on reference data and implicit neural expression according to an embodiment of the present invention expresses the new design (3D data) as design mapping data, which is a continuous value in the latent space, through deep learning, and uses the latent space Design mapping data can be used to provide creation, exploration, prediction, and optimization of new designs.

Figure 5 is a flowchart of a design creation method according to an embodiment of the present invention.

Referring to Figures 4 and 5, the design generation method (S10) according to an embodiment of the present invention includes a process of acquiring reference data (S100), a process of acquiring a new design by generative design (S200), Process of filtering new designs (S300), process of checking the number of new designs (S400), process of acquiring additional new designs by design generation model (S500), filtering of additional new designs and using the additional new designs as reference data It may include at least one of the process of repeating the above process (S600) and the process of evaluating a new design (S700).

The computing device 100 can generate a relatively large number of new designs using a small number of reference data through an optimized generative design method based on reference data and implicit neural expressions according to an embodiment of the present invention.

The optimized generative design method based on reference data and implicit neural representation according to an embodiment of the present invention can generate and evaluate a new design of a product using reference data. Here, the product can be any structure with a shape or appearance. Examples include cars, car wheels, car wheels, watches, smartphones, bridges, and buildings.

For convenience of explanation, the description in this specification is based on the wheels of a car. However, the scope of the present invention is not limited to automobile wheels, and can be applied to all products that can be designed.

In the S100 process, the computing device 100 may acquire reference data.

Computing device 100 may obtain reference data for a product. The computing device 100 may receive reference data through the communication interface 130 or obtain reference data stored in the storage 140.

Reference data may include the design of a product that is illustrated by a designer, designed by an engineer, and verified aesthetically and performance-wise. Alternatively, reference data refers to a technical blueprint containing the essential elements of a system, which may be copied by third parties. Third parties may improve or modify reference data as needed.

Reference data includes pattern images unrelated to the product (e.g., Pattern Img), images for two-dimensional products (e.g., Wheel Img), skeleton images for two-dimensional products (e.g., 2D Skel. Img), and images for three-dimensional products. Skeleton image for a 3D product (e.g. 3D Skel. Img), cross-sectional image for a 3D product (e.g. Cross Img), view image for a 3D product (e.g. View Img), black and white image for a product (e.g. Black Img) ), and may include at least one of a color image (e.g., Color Img) for the product.

Reference data may include 2D reference data or 3D reference data.

In this case, the 2D reference data is shape image data, pattern data, skeleton data, graph data, SDF data (distance information), depth data, implicit functions (function expressions or deep learning models), grammar, and multi-views obtained from each type of data. It may contain at least one of the data.

Additionally, 3D reference data includes CAD data, voxel data, mesh data, point cloud data, octree data, shape parameter data, pattern data, skeleton data, graph data, SDF data (distance information), and implicit functions (function expression or deep learning model). It may include at least one of:

Reference data can be called seed data because it is used as a seed to create a new design.

In the following specification, the term design or design regarding a product refers to data that can be expressed as input and output of the computing device 100.

Below, a process for obtaining reference data from an optimized generative design method based on reference data and implicit neural representations according to an embodiment of the present invention will be described.

Figure 6 is a flowchart of a process for obtaining reference data in an optimized generative design method based on reference data and implicit neural representations according to an embodiment of the present invention. Figures 7(a) to 7(d) show examples of preprocessed images generated by preprocessing input data.

Referring to FIG. 6, the process of acquiring reference data according to an embodiment of the present invention includes a process of acquiring input data (S210), a process of pre-processing the input data (S220), and a process of providing a pre-processed image (S230). It can be included. Here, the computing device 100 may provide a preprocessed image as reference data through a process of acquiring reference data.

However, the process of obtaining reference data according to an embodiment of the present invention is not limited to the above, and the reference data may include any type of data about the product. Additionally, reference data may also include data such as pattern images that are not related to the product.

In process S210, the computing device 100 may receive input data through the communication interface 130 or obtain input data stored in the storage 140. The input data may include at least one of an image for a two-dimensional product, an image for a three-dimensional product, and modeling for a three-dimensional product.

In process S220, the computing device 100 may preprocess the input data according to a preset preprocessing method. For example, preset preprocessing methods include a method of extracting a pattern image from input data, a method of extracting a skeleton image, a method of extracting a view image, a method of extracting a cross-sectional image, and a method of extracting a black-and-white image from a color image. It may include at least one of:

In process S221, referring to FIG. 7(a), the computing device 100 processes input data (e.g., an image for a two-dimensional product, an image for a three-dimensional product, and an image for a three-dimensional product) according to a preset preprocessing method. A pattern image can be extracted using a pattern extraction model (one of modeling methods).

In process S222, referring to FIG. 7(b), the computing device 100 processes input data (e.g., an image for a two-dimensional product, an image for a three-dimensional product, and an image for a three-dimensional product) according to a preset preprocessing method. You can extract a skeleton image for a two-dimensional product or a skeleton image for a three-dimensional product using a skeleton extraction model.

In process S223, referring to FIG. 7(c), the computing device 100 processes input data (e.g., an image for a two-dimensional product, an image for a three-dimensional product, and an image for a three-dimensional product) according to a preset preprocessing method. Modeling), you can use a view extraction model to extract a view image for a 3D product from a specific view. Here, the view image for a 3D product may refer to an image when the 3D product is viewed from a specific direction (view). Furthermore, the computing device 100 may extract a view image for a 3D product and also provide information about the specific view (e.g., [x,y,z]=[1,1,1]). .

In process S224, referring to FIG. 7(d), the computing device 100 processes input data (e.g., an image for a two-dimensional product, an image for a three-dimensional product, and an image for a three-dimensional product) according to a preset preprocessing method. Modeling), a cross-sectional image of a 3D product can be extracted at a certain depth according to a specific view and user input by using a cross-sectional extraction model.

Here, the cross-sectional image of the 3D product may mean an image of the cross-section of the 3D product at a depth according to the value input by the user while viewing the 3D product from a specific direction (view). Furthermore, the computing device 100 may extract a cross-sectional image of a three-dimensional product and provide information about the specific view and information about the depth of the pixel (e.g., [x,y,z]=[ 0,0,1]).

In process S225, the computing device 100 may extract a black-and-white image (e.g., binary image) for the product using a binary extraction model from input data (e.g., color image) according to a preset preprocessing method.

In process S220, the computing device 100 may perform one or two or more of the processes S221 to S225 in parallel with respect to the input data, respectively, and may perform a combination of two or more sequentially, through which preprocessing is performed. image can be obtained.

In process S230, the computing device 100 may provide the image preprocessed in process S100, and the preprocessed image may be used as reference data. In other words, the computing device 100 performs a preprocessing process on input data including at least one of an image for a two-dimensional product, an image for a three-dimensional product, and modeling for a three-dimensional product so that it can be used as reference data. Preprocessed images can be obtained.

Here, the preprocessed image is a pattern image (e.g., Pattern Img), an image for a two-dimensional product (e.g., Wheel Img), a skeleton image for a two-dimensional product (e.g., 2D Skel. Img), and an image for a three-dimensional product. Skeleton image (e.g., 3D Skel. Img), cross-sectional image of a three-dimensional product (e.g., Cross Img), view image of a three-dimensional product (e.g., View Img), black and white image of the product (e.g., Black Img) , may include at least one color image (e.g., Color Img) for the product, or a combination of two or more.

For example, the reference data may be a binary image of a two-dimensional product, or more specifically, a binary pixel image of a front view of the product converted to black and white binary. For example, the reference data may be a skeleton image, and more specifically, it may be an image obtained by extracting the skeleton of a two-dimensional (three-dimensional) product and converting it into a skeleton.

Referring again to FIG. 5, in the S200 process, the computing device 100 may acquire a new design through generative design. Specifically, the computing device 100 may generate a plurality of new designs by applying generative design based on reference data.

The computing device 100 may apply generative design to the reference data to obtain a plurality of new designs that have similarities to the reference data.

Generative design includes (i) a topology optimization methodology using at least one of the SIMP model, ESO model, BESO model, LevelSet model, MMC model, and deep learning model, (ii) dimension and shape optimization (Size/ At least one of (Shape Optimization) methodology, and (iii) Parametric Design methodology may be used.

The computing device 100 applies generative design to reference data based on input signals for the generative design methodology selected by the user or setting information for the generative design methodology preset for each product to create a plurality of new designs. It can be obtained by creating it.

There are two ways to create a generative design that resembles reference data. The first is a method of reflecting the similarity with the reference data in the objective function, and the second is a method of reflecting the similarity with the reference data in the sensitivity function that differentiates the objective function.

First, we will introduce the first method, a method of reflecting similarity in the objective function. The objective function of generative design can be composed of the sum of the performance objective function part (PERF), which corresponds to the matters that the user must satisfy during design, and the similarity part (SIM), which determines the similarity with reference data.

The performance objective function part (PERF) is an optimization problem related to product performance and may include factors such as product engineering performance, product manufacturability, and product aesthetics. In other words, it may mean requirements set by the user. The similarity objective function part (SIM) is a similarity optimization problem with respect to reference data, which may include the similarity between the new design and the reference data. The similarity objective function part is weighted ( ), the influence on the overall objective function of generative design can be adjusted.

We introduce the second method, a method of reflecting similarity in the sensitivity function. The sensitivity function may be composed of the sum of the differentiation of the performance objective function (PERF) corresponding to the requirements that the user must satisfy during design and the similarity with reference data.

The sensitivity function represents the rate of change of the objective function, and this value is repeatedly calculated to find the optimal design value. At this time, the more similar it is to the reference data (reference design), the greater the sensitivity, so that the new design (generated design) can be induced to resemble the reference data (reference design). At this time, the weight ( ), the influence of the reference data (reference design) of generative design can be adjusted.

In summary, the generative design of the present invention can create a new design similar to reference data by assigning similarity (SIM) to the objective function or similarity (SIM) to the sensitivity function, and the weight ( ), the degree of similarity can be adjusted. The present invention includes all generative design methods that modify the objective function or sensitivity function to create a generative design that resembles reference data.

In various embodiments, the computing device 100 may set an optimization zone to apply topology optimization to reference data. The optimization zone can be an entire region corresponding to a product in the reference data, or it can be a unit region that is repeated in the product. Alternatively, the computing device 100 may determine an optimization zone in the reference data based on at least one parameter (eg, symmetric equal division).

Here, the computing device 100 may extract an optimization zone from the reference data and apply topology optimization based on the optimization zone. Here, by performing topology optimization based on all or part of the reference data, a plurality of new designs similar to all or part of the reference data can be created.

Optimal design can be divided into shape optimization, which optimizes the shape or dimensions of the object, material optimization, which optimizes the material, and topology optimization, which optimizes the structure of the object. In particular, topology optimization is defined as a physics-based optimization problem for reference data that can satisfy the user's design requirements.

The topology optimization problem proposed in the present invention generates a plurality of new designs that maintain a similar shape to the reference data while having engineering performance similar to the reference data by introducing an objective function or sensitivity function including the design performance part and the similarity part. will be. The design performance mainly used in topology optimization is compliance, which is an indicator inversely proportional to structural rigidity, and the smaller the design, the better the performance. Topology-optimized design was developed as an answer to the question of how to distribute materials. Compliance minimization related to the rigidity of the product's structure is performed by redesigning the reference data.

As an example, when performing topology optimization on reference data for a car wheel, a user may change at least one parameter to create a plurality of new topology-optimized designs. At least one parameter can be specified by the user depending on the shape, performance, function, etc. of the product.

As an example, based on a car wheel, at least one parameter may include similarity weight, force ratio, volume ratio, symmetry equalization, etc. Similarity weight is a weight that determines how much a new design resembles reference data. Force Ratio is the ratio of the magnitude of the shear force to the vertical force of the wheel. Volume Ratio is the volume ratio of the new design compared to reference data. A symmetrical piece refers to a symmetrical piece of a wheel.

In various embodiments, the normal force expressed as tire pressure applied perpendicular to the automobile wheel wheel and the shear force expressed as the dotted line friction force of the wheel wheel are typical load conditions in optimization of two-dimensional automobile wheel design. is applied uniformly along the surface of the wheel. Force ratio of shear force to normal force is a user-specified parameter that can be changed when designing a new car wheel with topology optimization.

In various embodiments, the volume ratio of the new design compared to the reference data is a user-specified parameter that can be changed when a new design applies topology optimization to the reference data.

In various embodiments, the similarity ratio of the new design to the reference data is a user-specified parameter that can be changed when a new design applies topology optimization to the reference data, and recommended values may be automatically presented.

In various embodiments, the symmetric piece for the reference data is a user-specified parameter that can be changed when designing a new design that applies topology optimization to the reference data.

Meanwhile, the L1 Norm between the topology-optimized new design and the reference data indicates the similarity between both designs. On the other hand, cross-entropy can be applied instead of L1 Norm when the design of the product is related to a discrete problem, indicating the similarity between the new topology-optimized design and the reference data. On the other hand, the pixel-wise L1 Norm can be applied to the part that shows similarity between the topology-optimized new design and the reference data when the product design is related to a continuous problem.

Figure 8 shows various examples of new designs created using pattern images as reference data.

Referring to FIG. 8, in the S200 process of various embodiments, the computing device 100 uses the pattern image as reference data to perform generative design (e.g., topology optimization, dimension and shape optimization, parametric design) to create a new You can create a design.

In various embodiments, when applying a topology optimization methodology through generative design, the computing device 100 ensures the engineering performance of the product through the performance portion of the topology optimization objective function and resembles the pattern image through the similarity portion. You can create new designs for products. That is, the computing device 100 can create a new design for an aesthetically similar product by using the pattern image as reference data even if there is no reference data (for example, an initial wheel image) for the product to be newly designed.

Figure 9 shows various examples of new designs created using a skeleton image as reference data.

Referring to FIG. 9, in process S200 of various embodiments, the computing device 100 generates a skeleton image for a two-dimensional product (e.g., 2D Skel. Img) or a skeleton image for a three-dimensional product (e.g., 3D Skel. Img). You can create a new design by using generative design (e.g., topology optimization, dimension and shape optimization, parametric design) as reference data.

In various embodiments, when applying the topology optimization methodology with generative design, a skeleton image for a two-dimensional (3-dimensional) product is used instead of a general image for the two-dimensional (3-dimensional) product, thereby creating a two-dimensional (3-dimensional) product. (3D) The influence caused by the biased density (thickness, volume) in the general image of the product can be prevented from affecting the new design. Additionally, by using skeleton images for two-dimensional (three-dimensional) products, the new design can ensure that the density (thickness, volume) distribution for the product is not biased.

In various embodiments, the computing device 100 may create a new design for a product having a uniformly increased density (thickness, volume) according to a volume ratio, which is a user-specified parameter.

For example, a new design with a large volume ratio may provide a design for a product with relatively increased density (thickness, volume) compared to a new design with a small volume ratio.

Figure 10 shows various examples of new designs created using a view image or cross-sectional image as reference data.

Referring to the two diagrams at the top of FIG. 10 , computing device 100 may display a novel image resembling a view image in a particular direction in one of an image for a two-dimensional product, an image for a three-dimensional product, and modeling for a three-dimensional product. You can create a design.

Here, the computing device 100 can extract a view image (O1) in a specific direction using a view extraction model from one of an image for a two-dimensional product, an image for a three-dimensional product, and modeling for a three-dimensional product. there is. Additionally, the computing device 100 may also extract information about a view in a specific direction. Here, information about a specific direction view may represent the direction in which the product is viewed as a vector (e.g., [x, y, z]=[1,1,1]).

Furthermore, the computing device 100 may generate a new design (N1) by applying topology optimization to the view image (O1) in a specific direction as reference data. In this case, the new design may be a design similar to the view in the same direction as the specific direction view image.

Furthermore, the computing device 100 can extract a plurality of view images (O1, O2) for a plurality of view directions, and apply generative design (e.g., topology optimization, dimension and shape optimization, para. Multiple new designs (N1, N2) can be created by applying each metric design.

Here, the plurality of new designs may be topologically optimized designs corresponding to each of the plurality of view images. In this case, the objective function of the topology optimization is formed in plural numbers to have similarity parts corresponding to the plurality of view images, and by finding the answer to each of the objective functions of the plurality of topology optimizations, a plurality of new designs can be obtained.

Furthermore, the computing device 100 uses information about the plurality of view directions ([1,1,1], [1,0,1]) to combine a plurality of new designs corresponding to the plurality of view directions. You can create new images for two-dimensional products or new modeling for three-dimensional products.

In various embodiments, computing device 100 may extract a plurality of view images (O1, O2) for a plurality of view directions, and use the plurality of view images all together to perform generative design (e.g., topology optimization, dimension optimization, etc.). and shape optimization, parametric design) can be applied to create a new design. In this case, the objective function of topology optimization is formed by a plurality of similarity parts corresponding to multiple view images, and by finding the answer to the objective function of one topology optimization with a plurality of similarity parts (L1 distance term), a new You can get the design.

Meanwhile, referring to the lower drawing of FIG. 10, the computing device 100 responds to a user input from a view in a specific direction in one of an image for a two-dimensional product, an image for a three-dimensional product, and modeling for a three-dimensional product. A new design resembling a cross-sectional image can be created at a certain depth.

Here, the computing device 100 uses a cross-sectional extraction model from one of an image for a two-dimensional product, an image for a three-dimensional product, and modeling for a three-dimensional product to obtain a cross-sectional image (O3) of the product in a specific direction and at a predetermined depth. ) can be extracted. Here, the cross-sectional image of the product may mean an image of the cross-section of the product at a depth according to the value input by the user while viewing the two-dimensional or three-dimensional product from a specific direction (view).

Additionally, the computing device 100 may also extract information about a view in a specific direction and information about depth. Here, information about a specific direction view may represent the direction in which the product is viewed as a vector (e.g., [x, y, z]=[0,0,1]). Information about depth can be defined as 100% for the part of the product furthest from the user, and 0% for the part of the product closest to the user, assuming the total depth of the product based on a view in a specific direction is 100%. (e.g. depth = 50%).

Furthermore, the computing device 100 applies generative design (e.g., topology optimization, dimension and shape optimization, parametric design) to the cross-sectional image (O3) in a specific direction and at a predetermined depth as reference data to create a new design (N3). can be created. In this case, the new design may be a design similar to the cross-sectional image of the product in a specific direction and at a predetermined depth, with a view in the same direction and at a predetermined depth.

Furthermore, the computing device 100 can extract a plurality of cross-sectional images O3 for a specific direction and a plurality of depths, and apply generative design (e.g., topology optimization, dimension and shape optimization, para. By applying each metric design, multiple new designs (N3) can be created.

Here, the plurality of new designs may be topologically optimized designs corresponding to each of the plurality of cross-sectional images. In this case, the objective function of the topology optimization is formed in plural numbers to correspond to the plurality of cross-sectional images, and by finding the answer to each of the objective functions of the plurality of topology optimizations, a plurality of new designs can be obtained.

Furthermore, the computing device 100 uses information about a specific direction and multiple depths ([0, 0, 1]) to create a new image for a two-dimensional product by combining multiple new designs corresponding to multiple depths. Or, you can create new modeling for 3D products.

In various embodiments, the computing device 100 may extract a plurality of cross-sectional images O3 for a plurality of depth directions, and apply topology optimization to the plurality of cross-sectional images all together to create a new design. there is.

In this case, the objective function of topology optimization is formed by a plurality of similarity parts corresponding to a plurality of cross-sectional images, and by finding the answer to the objective function of one topology optimization with multiple similarity parts (L1 distance term), a new You can get the design.

In various embodiments, the computing device 100 uses multiple directional views of a product and view cross-sectional images extracted from multiple depths as reference data for generative design (e.g., topology optimization, dimension and shape optimization, parametric optimization). design) can be applied to create multiple new designs.

Referring to FIGS. 4 and 5 , in process S300, the computing device 100 may filter a plurality of new designs.

In the S300 process, the computing device 100 may perform at least one of a performance filtering process and a similarity filtering process of a new design.

In various embodiments, in the performance filtering process, the computing device 100 may evaluate new designs based on requirements set by the user at the time of design and exclude new designs that do not meet the requirements set by the user. It can be done. In this case, the new design that does not pass the requirements is called a low-performance new design.

In various embodiments, during the similarity filtering process, the computing device 100 may perform filtering to exclude similar designs among a plurality of new designs.

In various embodiments, computing device 100 may evaluate similarity in a high-dimensional domain in which the shape of the new design exists. Computing device 100 may perform an X spatial filtering process. Here, the computing device 100 measures the distance between two of the plurality of new designs in the This process can be performed for every new design. Here, the X space refers to a high-dimensional domain where the shape of the new design exists by itself. In other words, the X space refers to the new design itself created through generative design.

In various embodiments, the computing device 100 may evaluate similarity by mapping the shape of the new design onto a low-dimensional domain. Computing device 100 may perform a Z spatial filtering process. Here, the computing device 100 extracts low-dimensional feature values by projecting a plurality of new designs onto a latent space (Z-space) using an encoding model, measures the distance between the low-dimensional feature values of the two designs, and , if the distance between them is smaller than the preset second threshold, it can be viewed as a similar design and at least one of the two can be excluded. This process can be performed for every new design. Here, Z space refers to the space where the low-dimensional values of the new design are mapped.

In the S300 process, the computing device 100 may perform one or both of a performance filtering process and a similarity filtering process. Additionally, the computing device 100 may perform one or both of the above-described X-space filtering process and Z-space filtering process. However, the filtering process of the present invention is not limited to this, and various filtering methods that can classify similar designs can be applied.

Referring to FIGS. 4 and 5 , in process S400, the computing device 100 may check whether the number of filtered new designs exceeds a preset threshold. Here, the computing device 100 may perform the next S500 process if the number of filtered new designs does not exceed a preset threshold, and may perform the next S700 process if it exceeds the preset threshold.

In the S500 process, the computing device 100 may generate additional new designs by applying a design generation model to the new design.

The computing device 100 may learn a design creation model. As an example, the computing device 100 may learn a design generation model capable of generating a new additional new design using at least one of reference data and a new design.

Here, the learned design generation model may have a latent space (Z) learned by at least one of reference data and a new design. The latent space (Z) has a reduced dimension compared to the reference data and the new design, and the reference design and the new design can be expressed to be mapped on the reduced dimension (low dimension).

In various embodiments, the design creation model may be composed of an artificial intelligence model that extracts design data and maps it to the latent space (Z), and can create a new design using the latent space (Z).

The design generation model can be composed of any Generative Model using deep learning. As an example, the generative model is based on Vanilla GAN, DCGAN, cGAN, WGAN, EBGAN, BEGAN, CycleGAN, DiscoGAN, StarGAN, SRGAN, SEGAN, and SDF. It may consist of at least one of GAN, Shape GAN, and PolyGen models. However, the design generation model is not limited to the GAN model, and any generative model based on deep learning that can generate or interpolate new design data can be used. For example, the design generation model may include at least one of a deep learning model, a Boolean model, a morphing model, and an interpolation model.

The computing device 100 may generate a new additional design using the latent space (Z) of the learned generation model.

In the S600 process, the computing device 100 may perform filtering to exclude similar designs among additional new designs.

The filtering process described in the S300 process can be applied as is in the S600 process. Furthermore, the computing device 100 may perform the S100 process using additional new designs or filtered additional new designs as reference data. Here, the computing device 100 may perform a preprocessing process, such as extracting a pattern image or extracting a skeleton image, using an additional new design or a filtered additional new design.

The computing device 100 may repeat the S200 process of generating a new design by applying generative design using the additional new design and the filtered additional new design.

During S700, computing device 100 may evaluate a new design.

The computing device 100 may determine whether the new design performance obtained by analyzing the new design satisfies preset requirements. Requirements may be information entered by the user when creating a design, or may be information preset for each product.

Here, the new design performance is a first analysis result corresponding to the design requirements desired by the user, a second analysis result corresponding to the physical performance of the product corresponding to the design, and manufacturing, which is a matter considered when manufacturing a product corresponding to the design. It may include at least one of the third interpretation results corresponding to gender.

Correspondingly, the preset requirements are the first requirements corresponding to the preset design requirements, the second requirements corresponding to the preset physical performance that the product must have as a minimum, and the minimum considerations when manufacturing the product. It may include at least one of the third requirements corresponding to preset manufacturability.

In various embodiments, the computing device 100 may evaluate whether the requirement is satisfied by comparing at least one of the first to third analysis results of the new design with one of the corresponding first to third requirements.

The computing device 100 may repeatedly perform processes S100 to S600. Here, the computing device 100 may repeatedly perform processes S100 to S600 until the number of new designs filtered in process S400 is greater than a preset third threshold value.

Referring to FIGS. 4 and 5 , in process S800, the computing device 100 may perform a process of expressing a new design in a low-dimensional space.

Hereafter, for convenience of explanation, the new design input to the implicit neural expression model is referred to as the input design (Xi) and the output design (Xo).

In the S800 process, the computing device 100 may learn an implicit neural representation model using a plurality of new designs (Xi).

The implicit neural expression model can express multiple new designs (Xi) in a low-dimensional, continuous, parametric space (z), which can be referred to as multiple design mapping data (Z).

The computing device 100 may express a plurality of new designs (Xi) as continuous values in a low-dimensional space (z) using an implicit neural representation model.

The design mapping data (Z) has information about the new design (Xi) expressed in a low-dimensional space (z) that corresponds one-to-one with the new design (Xi), and the values between the plurality of design mapping data (Z) are continuous. Since it has , it is expressed in a differentiable way.

The computing device 100 uses a plurality of design mapping data (Z) in a low-dimensional continuous parametric space (z) to conduct a search process for a new design, an interpolation process for the design, a performance prediction process for the design, and a design process. At least one of the optimization process and reverse design process can be performed.

The computing device 100 is expressed on a continuous low-dimensional space (z) using an implicit neural expression model, and can store design mapping data (Z) corresponding to the new design (Xi) in storage ().

The new design (Xi) may include all seed data such as topology optimization, dimension and shape optimization, parametric design, and user data included in the generative design described above.

The computing device 100 maps all seed data to a low-dimensional latent space (z) through deep learning on an implicit neural expression model, and then processes the design mapping data (Z) within the latent space (z) to create new Design mapping data (Z') can be created.

The computing device 100 can generate an output design (Xo) that is distinct from the existing new design (Xi) using new design mapping data (Z') of the potential space (z), and the output design ( Since it is newly synthesized data, it can be called synthetic data.

Below, with reference to FIGS. 2 and 4 , the design search process, design interpolation process, design performance prediction process, design optimization process, and design reverse engineering process will be described.

In process S20, the computing device 100 searches for first design mapping data (Z ₁ ) among a plurality of design mapping data (Z) in a low-dimensional space (z) and corresponds to the first design mapping data (Z ₁ ). A design search process may be performed to obtain the first design (X ₁ ).

Because design mapping data (Z) expressed in a low-dimensional space (z) is expressed as continuous numerical values, it can be easier to explore than a new design (X) expressed as two-dimensional or three-dimensional design data. Therefore, in order to find a specific first design (X ₁ ), by searching the first design mapping data (Z ₁ ) corresponding to the first design (X ₁ ) in the low-dimensional space (z), the first design (X ₁ ) can be found.

After learning the implicit neural expression model as a new design (X), the computing device 100 can search for a desired new design in real time using the design mapping data (Z) expressed in the latent space (z).

Figure 11 shows an example of design interpolation in latent space.

Referring to FIG. 11, in process S20, the computing device 100 interpolates a plurality of second design mapping data (Z _{1 ) among a plurality of design mapping data (Z) in a low-dimensional space (z), and generates the interpolated first design mapping data (Z 1} ). 2 A design interpolation process may be performed to obtain a second design (X2) corresponding to the design mapping data (Z ₁ ').

In various embodiments, the computing device 100 selects N data from a plurality of design mapping data (Z) in a low-dimensional space (z) and interpolates them into second design mapping data (Z ₂ ) through an interpolation method, and interpolates A design interpolation process may be performed to obtain the second design (X2) corresponding to the second design mapping data (Z ₂ ).

In various embodiments, the computing device 100 applies an interpolation method (e.g., median value, average value, or average with different weights, etc.) to at least one design mapping data expressed in a low-dimensional space (z) to create a new Unique design mapping data can be created. Because this interpolation process takes place within a low-dimensional space (z), it can be performed with a simple operation.

In the conventional case, multiple new designs created through generative design cannot be expressed in a continuous design space. Therefore, it is not possible to create Design C, which is located in the middle of Design A and Design B. However, an embodiment of the present invention can express a new design by interpolating design mapping data corresponding to all designs in a low-dimensional latent space (z).

That is, after learning the implicit neural expression model as a new design (X), the computing device 100 can generate a desired new design in real time using the design mapping data (Z) expressed in the latent space (z).

In the S30 process, the computing device 100 generates a plurality of design mapping data (Z) in a low-dimensional space (z) and performance data (Y) of a plurality of designs (X) corresponding to the plurality of design mapping data (Z). Using the learned performance prediction model, a design prediction process can be performed to predict predicted performance data (Y ₃ ') through third design mapping data (Z ₃ ) for the third design (X ₃ ).

In other words, by learning a performance prediction model with design mapping data (Z) and performance data (Y) corresponding to design (X), the learned stratified prediction model can output predicted performance data with specific design mapping data as input. there is.

The computing device 100 may learn a performance prediction model using the design (X) and performance data (Y) corresponding to each design as learning data. In detail, a performance prediction model can be trained using the design mapping data (Z) and performance data (Y) of the design (X) expressed in a low dimension using a temporary neural expression model as training data.

In this case, the learned performance prediction model, together with the implicit neural representation model, can predict performance data (Y') for the new design (X). That is, the computing device 100 can input a new design (X) and predict the performance of a product of the new design using a performance prediction model.

In process S40, the computing device 100 generates fourth design mapping data (X ₄ ) corresponding to the fourth design ( A design optimization process can be performed to optimize Z ₄ ) and obtain optimized fourth design data (X ₄ ').

If the performance data of a specific design satisfies the user's desired requirements, the computing device 100 acquires design mapping data corresponding to the corresponding design, and generates fourth design mapping data (Z ₄₎ based on the design mapping data. ) and obtaining the optimized fourth design (X ₄ '), it is possible to obtain the fourth design (X ₄ ') optimized to meet the user's desired requirements.

In various embodiments, the computing device 100 may perform topology optimization by applying a topology optimization technique to a plurality of design mapping data in a low-dimensional space (z).

Conventional automatic parameterization optimal design technology requires the designer to perform parameterization based on the base design and define design variables to be optimized in order to optimize the 3D shape. Because of this, the optimal design can only be found within a local design space. However, in the embodiment of the present invention, an optimal design can be performed in the original design space because the 3D shape is automatically parameterized in the latent space.

In process S40, the computing device 100 uses a reverse engineering model learned with performance data (Y) of a low-dimensional space (z) and a plurality of design mapping data (Z) corresponding to the performance data (Y), 5 Obtain the fifth design mapping data (Z ₅ ) corresponding to the performance data (Y ₅ ), and perform a reverse engineering process to obtain the fifth design (X ₅ ) corresponding to the fifth design mapping data (Z ₅ ). can do.

Here, the reverse engineering model can output a design (X) that satisfies the requirements when the user inputs performance data (Y) for the desired requirements. By receiving desired performance specifications for a specific product, the computing device 100 may output a design for a specific product that satisfies the input performance specifications. In other words, users can reverse engineer the design of a product as long as they know the product's performance data.

All of the above-mentioned models may refer to artificial intelligence models. Here, the artificial intelligence model is composed of one or more network functions, and the one or more network functions may be composed of a set of interconnected computational units that can be generally referred to as 'nodes'. These 'nodes' may also be referred to as 'neurons'. One or more network functions are composed of at least one or more nodes. Nodes (or neurons) that make up one or more network functions may be interconnected by one or more 'links'.

Within an artificial intelligence model, one or more nodes connected through a link can form a relative input node and output node relationship. The concepts of input node and output node are relative, and any node in an output node relationship with one node may be in an input node relationship with another node, and vice versa. As described above, input node to output node relationships can be created around links. One or more output nodes can be connected to one input node through a link, and vice versa.

In a relationship between an input node and an output node connected through one link, the value of the output node may be determined based on data input to the input node. Here, the nodes connecting the input node and the output node may have a weight. Weights may be variable and may be varied by the user or algorithm in order for the artificial intelligence model to perform the desired function. For example, when one or more input nodes are connected to one output node by respective links, the output node is set to the values input to the input nodes connected to the output node and the links corresponding to each input node. The output node value can be determined based on the weight.

As described above, in an artificial intelligence model, one or more nodes are interconnected through one or more links to form an input node and output node relationship within the artificial intelligence model. The characteristics of the artificial intelligence model can be determined according to the number of nodes and links within the artificial intelligence model, the correlation between nodes and links, and the weight value assigned to each link. For example, if there are two artificial intelligence models with the same number of nodes and links and different weight values between the links, the two artificial intelligence models may be recognized as different from each other.

Some of the nodes constituting the artificial intelligence model may form one layer based on the distances from the initial input node. For example, a set of nodes with a distance n from the initial input node may constitute n layers. The distance from the initial input node can be defined by the minimum number of links that must be passed to reach the node from the initial input node. However, this definition of the layer is arbitrary for explanation purposes, and the order of the layer within the artificial intelligence model may be defined in a different way from that described above. For example, a layer of nodes may be defined by distance from the final output node.

The initial input node may refer to one or more nodes in the artificial intelligence model into which data is directly input without going through links in relationships with other nodes. Alternatively, in an artificial intelligence model network, in the relationship between nodes based on links, it may mean nodes that do not have other input nodes connected by links. Similarly, the final output node may refer to one or more nodes that do not have an output node in their relationship with other nodes among the nodes in the artificial intelligence model. Additionally, hidden nodes may refer to nodes that constitute an artificial intelligence model rather than the first input node and the last output node. The artificial intelligence model according to an embodiment of the present invention may have more nodes in the input layer than the nodes in the hidden layer close to the output layer, and is an artificial intelligence model in which the number of nodes decreases as it progresses from the input layer to the hidden layer. You can.

An artificial intelligence model may include one or more hidden layers. The hidden node of the hidden layer can take the output of the previous layer and the output of surrounding hidden nodes as input. The number of hidden nodes for each hidden layer may be the same or different. The number of nodes in the input layer may be determined based on the number of data fields of the input data and may be the same as or different from the number of hidden nodes. Input data input to the input layer can be operated by the hidden node of the hidden layer and output by the fully connected layer (FCL), which is the output layer.

The computing device 100 can learn an artificial intelligence model using learning data. The computing device 100 may perform learning on one or more network functions constituting an artificial intelligence model using a learning data set.

In one embodiment, the computing device 100 inputs each set of learning input data into one or more network functions, extracts output data values calculated by one or more network functions as feature values, and artificial intelligence based on the extracted feature values. Intelligence models can be trained.

In another embodiment, the computing device 100 inputs each of the learning input data sets into one or more network functions, and sets each output data calculated by one or more network functions and a learning output data set corresponding to the label of each of the learning input data sets. By comparing each, the error can be derived.

In the training of an artificial intelligence model, learning input data may be input to the input layer of one or more network functions, and learning output data may be compared with the output of one or more network functions. The computing device 100 may train an artificial intelligence model based on the calculation results of one or more network functions on the learning input data and the error of the learning output data (label).

Additionally, the computing device 100 may adjust the weight of one or more network functions based on the error using a backpropagation method. That is, the computing device 100 may adjust the weights so that the output of one or more network functions is closer to the learning output data based on the calculation results of one or more network functions on the learning input data and the error of the learning output data.

When learning of one or more network functions has been performed for more than a predetermined epoch, the computing device 100 may determine whether to stop learning using verification data. The predetermined epoch may be part of the overall learning target epoch. Verification data may consist of at least part of a labeled training data set. That is, the computing device 100 performs learning of the artificial intelligence model through a learning data set, and after learning of the artificial intelligence model is repeated for more than a predetermined epoch, the learning effect of the artificial intelligence model is predetermined using verification data. It is possible to determine whether it is above a determined level. For example, when the computing device 100 performs learning with a target repetition learning number of 10 using 100 learning data, it performs 10 repetition learning, which is a predetermined epoch, and then uses 10 verification data. By performing three iterative learning, if the change in the artificial intelligence model output during the three iterative learning is below a predetermined level, it is determined that further learning is meaningless and learning can be terminated. In other words, the verification data can be used to determine the completion of learning based on whether the effect of learning for each epoch is above or below a certain level in the iterative learning of the artificial intelligence model. The number of training data, verification data, and number of repetitions described above are only examples and are not limited thereto.

The computing device 100 may generate an artificial intelligence model by testing the performance of one or more network functions using a test data set and determining whether to activate one or more network functions. Test data may be used to verify the performance of an artificial intelligence model and may consist of at least part of a learning data set. For example, 70% of the training data set can be used for training the artificial intelligence model (i.e. learning to adjust the weights to output similar results as labels), and 30% can be used to improve the performance of the artificial intelligence model. It can be used as test data for verification.

The computing device 100 may input a test data set to a trained artificial intelligence model, measure the error, and determine whether to activate the artificial intelligence model depending on whether it exceeds a predetermined performance. The computing device 100 verifies the performance of the trained artificial intelligence model by using test data on the trained artificial intelligence model, and if the performance of the trained artificial intelligence model is higher than a predetermined standard, the artificial intelligence model can be used in another application. You can enable it for use.

Additionally, the artificial intelligence model may be learned using at least one of supervised learning, unsupervised learning, and semi-supervised learning. Learning of artificial intelligence models is intended to minimize errors in output. In the learning of an artificial intelligence model, learning data is repeatedly input into the artificial intelligence model, the output of the artificial intelligence model and the error of the target for the learning data are calculated, and the error of the artificial intelligence model is calculated to reduce the error. This is the process of updating the weight of each node of the artificial intelligence model by backpropagating from the output layer to the input layer.

In the case of teacher learning, learning data in which the correct answer is labeled for each learning data is used (i.e., labeled learning data), and in the case of non-teacher learning, the correct answer may not be labeled in each learning data. That is, for example, in the case of teacher learning regarding data classification, the learning data may be data in which each learning data is labeled with a category. Learning data is input to the artificial intelligence model, and error can be calculated by comparing the output (category) of the artificial intelligence model and the label of the learning data.

As another example, in the case of non-teachable learning for data classification, the error can be calculated by comparing the input learning data with the artificial intelligence model output. The calculated error is back-propagated in the artificial intelligence model in the reverse direction (i.e., from the output layer to the input layer), and the connection weight of each node in each layer of the artificial intelligence model can be updated according to back-propagation.

The amount of change in the connection weight of each updated node may be determined according to the learning rate. The calculation of an artificial intelligence model on input data and backpropagation of errors can constitute a learning cycle (epoch). The learning rate can be applied differently depending on the number of repetitions of the learning cycle of the artificial intelligence model. For example, in the early stages of training an artificial intelligence model, a high learning rate can be used to increase efficiency by allowing the artificial intelligence model to quickly achieve a certain level of performance, and in the latter stages of training, a low learning rate can be used to increase accuracy.

In the training of an artificial intelligence model, generally, the learning data can be a subset of real data (i.e., the data that you want to process using the learned artificial intelligence model), and therefore, the error for the learning data is reduced, but for the real data There may be a learning cycle in which errors increase. Overfitting is a phenomenon in which errors in actual data increase due to excessive learning on training data. Overfitting can cause errors in machine learning algorithms to increase. To prevent such overfitting, various optimization methods can be used. To prevent overfitting, methods such as increasing the learning data, regularization, or dropout, which omits some of the network nodes during the learning process, can be applied.

In various embodiments, an artificial intelligence model may include a plurality of artificial intelligence models. Multiple artificial intelligence models include machine learning models such as random forest model and support vector machine (SVM), deep learning models such as convolutional neural network (CNN) series models, and circulation models. It may include at least one of the neural network (RNN, Recurrent Neural Network) series models, and may be composed of the PointNet series and 3D CNN series specialized for 3D deep learning and a graphic neural network (Graph Neural Network, GNN), and may include multiple models. You can also do an ensemble. Additionally, all models can utilize both 2D and 3D data.

The present invention has been described with reference to the embodiments shown in the drawings, but these are merely exemplary, and those skilled in the art will understand that various modifications and other equivalent embodiments are possible therefrom. Therefore, the true scope of technical protection of the present invention should be determined by the technical spirit of the attached registration claims.

100: computing device

Claims (17)

In a generative design method performed by a computing device,
A generative design method comprising the process of applying generative design to reference data to obtain a plurality of new designs that have similarities to the reference data. According to paragraph 1,
A process of filtering the plurality of new designs;
A process of obtaining a plurality of additional new designs generated using a design generation model based on the plurality of filtered new designs;
a process of filtering the plurality of additional new designs; and
A generative design method further comprising obtaining the plurality of new designs by applying the generative design using the filtered plurality of additional new designs as the plurality of reference data. According to paragraph 1,
Further comprising the process of obtaining the reference data,
The first reference data includes 2D reference data or 3D reference data,
The 2D reference data includes at least one of shape image data, pattern data, skeleton data, graph data, SDF data (distance information), depth data, implicit function data, grammar, and multi-view data obtained from each type of data,
The 3D reference data includes CAD data, voxel data, mesh data, point cloud data, octree data, shape parameter data, pattern data, skeleton data, graph data, SDF data (distance information), and implicit functions (function expression or deep learning model). A generative design method, including at least one. According to paragraph 1,
The generative design is
(i) Topology optimization methodology using at least one of SIMP, ESO, BESO, LevelSet, MMC and deep learning,
(ii) dimensional and shape optimization methodologies, and
(iii) A generative design method, using at least one of the parametric design methodologies. According to paragraph 1,
(i) the objective function of the generative design,
It consists of a performance function part corresponding to the requirements entered by the user and a similarity function part that determines similarity to the reference data, or
(ii) the sensitivity function of the generative design above,
Consisting of a part that differentiates a performance function corresponding to the requirements entered by the user and a similarity function part that determines similarity to the reference data,
Generative design method According to paragraph 2,
The process of filtering the plurality of new designs or additional new designs is,
It includes at least one of a performance filtering process and a similarity filtering process of the new design,
The performance filtering process filters low-performance new designs through evaluation of requirements set by the user,
The similarity filtering process is a generative design method in which similarity is evaluated and filtered on a high-dimensional domain where the shape of the new design exists, or similarity is evaluated and filtered by mapping the shape of the new design onto a low-dimensional domain. According to paragraph 2,
The design generation model is a generative design method including at least one of a deep learning model, a Boolean model, a morphing model, and an interpolation model. According to paragraph 1,
A process of learning an implicit neural expression model using the plurality of new designs and mapping the plurality of new designs into a low-dimensional, continuous, parametric space (z) to obtain a plurality of design mapping data; and
Using the plurality of design mapping data in the low-dimensional, continuous, parametric space (z), it includes at least one of a design exploration process, a design interpolation process, a design performance prediction process, a design optimization process, and a design reverse engineering process. A generative design method. According to clause 8,
In the design optimization process,
A generative design method that performs topology optimization by applying a topology optimization technique to the plurality of design mapping data in the low-dimensional, continuous, parametric space (z). processor;
network interface;
Memory; and
Includes a computer program loaded into the memory and executed by the processor,
The computer program is,
A generative design device comprising instructions for applying generative design to reference data for a product to obtain a plurality of new designs having similarity to the reference data. Combined with a computing device,
A generative design computer program stored in a computer-readable recording medium for executing a process of applying generative design to reference data for a product to obtain a plurality of new designs having similarities to the reference data. In a method performed by a computing device,
A process of acquiring multiple design mapping data (Z) by mapping multiple designs (X) for a product into a low-dimensional latent space (z) using an implicit neural expression model; and
A generative design method comprising at least one of a design exploration process, a design interpolation process, a design performance prediction process, a design optimization process, and a design reverse engineering process using the plurality of design mapping data in the latent space (z). According to clause 12,
The design search process searches for first design mapping data (Z ₁ ) among the plurality of design mapping data (Z) in the latent space (z), and searches for first design mapping data (Z 1 ) corresponding to the first design mapping data (Z ₁ ) in the latent space (z). A generative design method that obtains 1 design (X ₁ ). According to clause 12,
The design interpolation process selects N data from the plurality of design mapping data (Z) in the latent space (z) and interpolates them into second design mapping data (Z ₂ ) through an interpolation method, and the interpolated second design mapping data (Z 2 ) is selected from the latent space (z). A generative design method, obtaining a second design (X ₂ ) corresponding to design mapping data (Z ₂ ). According to clause 12,
The design prediction process is learned with the plurality of design mapping data (Z) of the latent space (z) and the performance data (Y) of the plurality of designs (X) corresponding to the plurality of design mapping data (Z). A generative design method that predicts predicted performance data (Y ₃ ') through third design mapping data (Z ₃ ) for a third design (X ₃ ) using a performance prediction model. According to clause 12,
The design optimization process is based on the performance data (Y) of the plurality of designs (X), and fourth design mapping data (Z ₄ ) corresponding to the fourth design (X ₄ ) within the latent space (z). A generative design method that optimizes and obtains an optimized fourth design (X ₄ '). According to clause 12,
The reverse design process uses a reverse design model learned with the performance data (Y) of the latent space (z) and the plurality of design mapping data (Z) corresponding to the performance data (Y), and a fifth performance A generative design method, obtaining fifth design mapping data (Z ₅ ) corresponding to data (Y ₅ ), and obtaining fifth design data (X ₅ ) corresponding to the fifth design mapping data (Z ₅ ). .