Material designing device, material designing method, and program

Abstract

This material designing device is provided with: a search unit (121) that sets a generation condition (21) for generating a three-dimensional structure (23) of a material, the generation condition (21) including information indicating at least one of the characteristic and structure of the material; a cross-sectional image generation unit (123) that generates a cross-sectional image (22) of the material on the basis of the generation condition (21) set by the search unit (121); and a three-dimensional structure generation unit (124) that generates the three-dimensional structure (23) of the material on the basis of the cross-sectional image (22) generated by the cross-sectional image generation unit (123).

Description

Material Design Device, Material Design Method, and Program

[0001] Aspects of the present disclosure relate to a material design device, a material design method, and a program.

[0002] Material design plays an important role in modern science and technology. The discovery and improvement of new materials promote innovation in various industries. However, current material design processes often rely mainly on experiments based on the operator's rules of thumb and intuition.

[0003] In experimental-based design, since it is necessary to repeat a large number of prototypes and evaluations, a lot of man-hours are required to find appropriate design conditions. Furthermore, since there are limitations to the design variables that can be explored by experiments, it is easy to fall into local optimal designs.

[0004] In recent years, in response to these problems, highly efficient and wide-ranging material design methods using sampling from probabilistic models such as black-box optimization and generative models have been proposed.

[0005] For example, in Patent Document 1, a method is proposed in which a crystal structure is represented by lattice vectors and atomic coordinates, and a novel structure is searched for using a genetic algorithm.

[0006] Alternatively, in Patent Document 2, a method is proposed in which the three-dimensional structure of a protein is represented by an amino acid sequence and a map representing the distances between them, and a novel structure is searched for using a generative model.

[0007] Alternatively, in Non-Patent Document 1, a method is presented in which the microstructure of an electrode material is represented by a three-dimensional structure in voxel format, and a novel structure is searched for using a generative model.

[0008] Japanese Patent Application Laid-Open No. 2018-010428, Japanese Patent Application Laid-Open No. 2024-506535

[0009] Hsu T, 6 others, “Microstructure Generation via Generative Adversarial Network for Heterogeneous, Topologically Complex 3D Materials”, Augmenting Physics-based Models in ICME with Machine Learning and Uncertainty Quantification, Volume 73, pages 90-402, [online], December 2, 2020, JOM, [October 30, 2024], Internet <URL: https: / / doi.org / 404007 / s11838020-04484-y>, English

[0010] However, the method disclosed in Patent Document 1 is only applicable to materials with a periodic structure, such as crystalline materials. Similarly, the method disclosed in Patent Document 2 is only applicable to materials with a series structure, such as low molecular weight compounds or proteins. Therefore, the methods disclosed in Patent Documents 1 and 2 cannot be used for materials that do not have a periodic or series structure, such as particle-filled composite materials.

[0011] Furthermore, the method disclosed in Non-Patent Document 1 represents the material structure as three-dimensional voxel data, so it can be applied even if the material does not have a periodic or sequential structure. However, because the three-dimensional structure is generated directly from the generative model, there are challenges in terms of the cost and stability of structure generation.

[0012] In this disclosure, a material having a sequential structure is defined as a material having a sequential structure, which is a one-dimensional data structure such as a string, represented by SMILES (Simplified Molecular Input Line Entry System) or an amino acid sequence.

[0013] Furthermore, the materials that do not have a periodic or sequential structure, as covered by the material design method of this disclosure, include, but are not limited to, particle-filled composite materials consisting of particles or fillers of multiple material types. For example, the material design method of this disclosure may also cover microstructured materials such as ceramics, amorphous materials, and polycrystalline materials.

[0014] This disclosure provides a material design apparatus, a material design method, and a program that can more efficiently generate the three-dimensional structure of a material.

[0015] A material design apparatus according to one aspect of the present disclosure includes: a search unit that sets generation conditions for generating a three-dimensional structure of a material, which includes information indicating at least one of the material's properties and structure; a cross-sectional image generation unit that generates a cross-sectional image of the material based on the generation conditions set by the search unit; and a three-dimensional structure generation unit that generates a three-dimensional structure of the material based on the cross-sectional image generated by the cross-sectional image generation unit.

[0016] Furthermore, a material design method according to one aspect of the present disclosure is a material design method performed by a computer, comprising: a search step of setting generation conditions for generating a three-dimensional structure of a material including information indicating at least one of the material's properties and structure; a cross-sectional image generation step of generating a cross-sectional image of the material based on the generation conditions set by the search step; and a three-dimensional structure generation step of generating a three-dimensional structure of the material based on the cross-sectional image generated by the cross-sectional image generation step.

[0017] Furthermore, one aspect of this disclosure can be implemented as a program that causes a computer to execute the above-mentioned material design method. Alternatively, one aspect of this disclosure can be implemented as a computer-readable non-temporary recording medium that stores the program.

[0018] These comprehensive or specific embodiments may be implemented as systems, devices, methods, integrated circuits, computer programs, or recording media, or as any combination of systems, devices, methods, integrated circuits, computer programs, and recording media.

[0019] According to one aspect of this disclosure, a material design apparatus, etc., can be used to design materials more efficiently.

[0020] Figure 1 is a block diagram showing the configuration of the design system according to the embodiment. Figure 2 is a diagram illustrating the overview of the design system according to the embodiment. Figure 3 is a block diagram showing the functional configuration of the learning unit according to the embodiment. Figure 4 is a block diagram showing the functional configuration of the design unit according to the embodiment. Figure 5 is a flowchart illustrating the operation of the design unit according to the embodiment. Figure 6 is a flowchart illustrating the operation of the design unit in S12 of Figure 5. Figure 7 is a diagram showing an example of a display screen shown on the UI unit according to the embodiment. Figure 8 is a diagram showing another example of a display screen shown on the UI unit according to the embodiment. Figure 9 is a diagram showing yet another example of a display screen shown on the UI unit according to the embodiment.

[0021] (Summary of the Disclosure) Embodiments of the Disclosure will be described in detail below with reference to the drawings as appropriate. However, descriptions that are unnecessarily detailed may be omitted. For example, detailed descriptions of already well-known matters and redundant descriptions of substantially identical configurations may be omitted. This is to avoid the following description becoming unnecessarily verbose and to facilitate understanding by those skilled in the art.

[0022] The embodiments described below are examples only, and each is either a comprehensive or specific example. The numerical values, shapes, materials, components, arrangement and connection configurations of components, steps, and the order of steps shown in the following embodiments are examples only and are not intended to limit this disclosure. Furthermore, any components among the components of the following embodiments that are not described in an independent claim will be described as optional components. In other words, this disclosure is not limited to the following embodiments.

[0023] Further advantages and effects of one aspect of this disclosure will be made apparent from the specification and drawings. Such advantages and / or effects are provided by several embodiments and features described in the specification and drawings, but not all of them are necessarily provided in order to obtain one or more identical features.

[0024] The attached drawings and the following description are provided to enable a person skilled in the art to fully understand this disclosure, and are not intended to limit the subject matter described in the claims.

[0025] For example, each figure is a schematic diagram and not necessarily a strictly accurate representation. Therefore, for example, the scale may not necessarily match in each figure. Also, in each figure, substantially identical components are given the same reference numerals, and redundant explanations are omitted or simplified.

[0026] Furthermore, in this specification, ordinal numbers such as "first," "second," etc., do not mean the number or order of components unless otherwise specified, but are used to avoid confusion and to distinguish similar components.

[0027] (Embodiment) [Configuration] First, the configuration of the design system according to the embodiment will be described. Figure 1 is a block diagram showing the configuration of the design system 10 according to the embodiment. Figure 1 shows an example of the hardware configuration of the design system 10.

[0028] The design system 10 is a system capable of designing materials. Designing a material means exploring the three-dimensional structure of the material. Based on information indicating material properties or material structure, the design system 10 can generate a three-dimensional structure of a material having said material properties or material structure. Generating a three-dimensional structure of a material means generating information indicating the three-dimensional structure of the material. Information indicating the three-dimensional structure of a material is, for example, three-dimensional point cloud data.

[0029] As shown in Figure 1, the design system 10 comprises a storage device 11, a processing device 12, a UI (User Interface) device 13, and a communication device 14, which are interconnected via a bus or network 15.

[0030] The programs or instructions that implement the functions and processes described later in the design system 10 are stored, for example, in the storage device 11. Alternatively, these programs or instructions may be downloaded from some external device (e.g., a server) via the network 15 or the like, or provided from a removable storage medium such as flash memory.

[0031] The storage device 11 is a storage device that stores information used for the operation of the design system 10, which executes the functions of the DB unit 110 described later. The storage device 11 can be implemented by, for example, RAM (Random Access Memory), flash memory, a hard disk drive, etc. The storage device 11 stores installed programs or instructions, as well as files, data, etc., used to execute the programs or instructions. The storage device 11 may be implemented by a non-transitor storage medium.

[0032] The processing unit 12 is an information processing device that executes the processing of the learning unit and the design unit, which will be described later. The processing unit 12 may be implemented by, for example, a general-purpose processor or controller (circuit), or by a dedicated processor or controller (circuit). When the processing unit 12 is implemented by a general-purpose processor, it may be implemented by one or more CPUs (Central Processing Units), GPUs (Graphics Processing Units), processing circuits (Processing Circuits), or FPGAs (Field Programmable Gate Arrays), which may consist of one or more processor cores. The processing unit 12 executes the functions and processing of the design system 10 according to the program or instructions stored in the storage device 11, data such as parameters used to execute the program or instructions, etc.

[0033] The UI device 13 is a device that performs the functions of the UI section described later. The UI device 13 is implemented by, for example, an input device such as a keyboard and a mouse, and an output device such as a display. The input device that implements the UI device 13 may be a camera or a microphone, and the output device that implements the UI device 13 may be a speaker, a headset, or a printer. The UI device 13 may also be implemented by an input / output device such as a touch panel. The UI device 13 implements an interface between the user and the design system 10. For example, the user operates the design system 10 by operating the GUI (Graphical User Interface) displayed on a display screen such as a display or touch panel using a keyboard, mouse, etc. The UI device 13 includes a display unit and an operation reception unit. The display unit is, for example, a display screen on which various images are displayed, and is implemented by the output device described above. The operation reception unit is, for example, a device that receives user input, and is implemented by the input device described above.

[0034] The communication device 14 is a communication circuit that performs communication between various devices. The communication device 14 is implemented by a communication circuit that performs communication processing with a communication network such as a LAN (Local Area Network). The communication device 14 may also be implemented by various communication circuits that perform communication processing with external devices, the Internet, and communication networks such as a VPN (Virtual Private Network).

[0035] The hardware configuration of the design system 10 described above is merely an example. The design system 10 described herein may be implemented using other appropriate hardware configurations.

[0036] Next, an overview of the design system 10 according to the embodiment will be described. Figure 2 is a diagram illustrating the overview of the design system 10 according to the embodiment. Figure 2 shows an example of data input and output performed by the design system 10 in the embodiment.

[0037] As shown in Figure 2, the design system 10 inputs the generation conditions 21 into the cross-sectional image generation model 31 and obtains a cross-sectional image 22 as output. Furthermore, the design system 10 inputs the cross-sectional image 22 into the three-dimensional structure generation model 32 and obtains three-dimensional structures 23 and 24 as output.

[0038] The generation condition 21 is a condition that serves as input to the cross-sectional image generation model 31. The generation condition 21 is a condition for generating the three-dimensional structure of the material. The generation condition 21 includes information indicating at least one of the material's properties and structure. For example, the generation condition 21 is vector data consisting of values ​​of material properties such as dielectric constant, thermal conductivity, and elastic modulus. In other words, the generation condition 21 includes information indicating the material's properties. Alternatively, for example, the generation condition 21 is vector data consisting of values ​​representing geometric features of the material structure, such as the surface area of ​​particles, the distance distribution between particles, and the packing density. In other words, the generation condition 21 includes information indicating the material's structure. Furthermore, the generation condition 21 may be a vector consisting of latent variables randomly sampled from a probability distribution, which can be used as input to generative models such as GANs (Generative Adversarial Networks) or VAEs (Variational Autoencoders). These latent variables do not directly represent the geometric features of the material structure, but indirectly represent the geometric features of the material structure as a low-dimensional representation of the material structure. Furthermore, the generation condition 21 may be a vector consisting of any combination of the above material properties, the geometric features of the material structure, and the above latent variables.

[0039] The cross-sectional image 22 is the output of the cross-sectional image generation model 31 and the input of the three-dimensional structure generation model 32. In other words, the cross-sectional image generation model 31 outputs the cross-sectional image 22 based on the input generation conditions 21. The cross-sectional image 22 is, for example, a cross-sectional image of a material structure. The cross-sectional image 22 is an image in which the material structure is represented by encoding the type of material into the value of each pixel of the image. Alternatively, the cross-sectional image 22 is an image in which the material structure is represented by encoding physical properties such as dielectric constant and thermal conductivity into the value of each pixel of the image. Alternatively, the cross-sectional image 22 is an image in which the material structure is represented by encoding the type of material into each channel of the image. The cross-sectional image 22 may be data measured using, for example, SEM (Scanning Electron Microscope), or data generated using simulation methods such as DEM (Discrete Element Method).

[0040] The three-dimensional structures 23 and 24 are outputs of the three-dimensional structure generation model 32. In other words, the three-dimensional structure generation model 32 outputs the three-dimensional structures 23 and 24 based on the input cross-sectional image 22. The three-dimensional structures 23 and 24 are 3D data that captures the characteristics of the material structure, and are, for example, voxel data, point cloud data, or surface data. The three-dimensional structures 23 and 24 may be data measured using FIB (Focused Ion Beam)-SEM, or data generated using simulation methods such as DEM.

[0041] There is a one-to-one correspondence between the generation condition 21 and the cross-sectional image 22. That is, the cross-sectional image generation model 31, which takes the generation condition 21 as input and the cross-sectional image 22 as output, is a deterministic model, such as the Generator of GANs. A deterministic model is a model in which there is a one-to-one correspondence between input and output. In other words, a deterministic model is a model that generates the same cross-sectional image 22 when the same generation condition 21 is input.

[0042] On the other hand, the cross-sectional image 22 and the three-dimensional structure 23 have a one-to-many correspondence. That is, for one cross-sectional image 22, a plurality of three-dimensional structures (including the three-dimensional structure 23 and the three-dimensional structure 24) correspond. That is, the three-dimensional structure generation model 32 that takes the cross-sectional image 22 as input and the three-dimensional structure 23 as output is, for example, a probabilistic model such as a diffusion model or a VAE. A probabilistic model is a model in which the input and output have a one-to-many correspondence. That is, a probabilistic model is a model that can generate different three-dimensional structures 23 when the same cross-sectional image 22 is input.

[0043] Note that the generation condition 21 and the cross-sectional image 22 may have a one-to-many correspondence, or the cross-sectional image 22 and the three-dimensional structure 23 may have a one-to-one correspondence.

[0044] [Configuration of Learning Unit] The design system 10 according to the embodiment can perform material design using a learned learning model, but the design system 10 may also be able to learn the learning model. Hereinafter, the configuration of the learning unit for learning the learning model according to the embodiment will be described while referring to FIG. 3. FIG. 3 is a block diagram showing the functional configuration of the learning unit 200 according to the embodiment. The processing by the learning unit 200 is usually performed before the processing by the design unit described later, but the processing by the learning unit 200 may be performed again using the design result by the design unit. As shown in FIG. 3, the design system 10 includes a DB unit 110, a UI unit 130, and a learning unit 200. The learning unit 200 includes a cross-sectional image characteristic prediction model learning unit 201, a cross-sectional image generation model learning unit 202, a three-dimensional structure characteristic prediction model learning unit 203, and a three-dimensional structure generation model learning unit 204. Further, the learning unit 200 can acquire a dataset of virtual cross-sectional images 205, measured cross-sectional images 206, virtual three-dimensional structures 207, and measured three-dimensional structures 208. The learning unit 200 may include a storage unit in which the dataset is stored.

[0045] The UI unit 130 receives settings for model learning and data preprocessing from the user and inputs them to the learning unit 200. Also, the UI unit 130 displays the learning results of the model output from the learning unit 200 to the user.

[0046] The DB unit 110 stores learning data and evaluation data. The learning unit 200 acquires the learning data and evaluation data stored in the DB unit 110. Also, the DB unit 110 stores the model output by the learning unit 200. In the DB unit 110, as will be described in more detail below, material structures, experimental data and simulation data regarding material properties associated with the material structures, and the history of parameters set in the past, etc. are stored. Also, the DB unit 110 may store information regarding simulation models used in the past, and generation models or prediction models, etc.

[0047] The virtual cross-sectional image 205 is a data set including 2D data showing a cross-section of a material and data associated with material properties of the material having the cross-section shown by the 2D data. The virtual cross-sectional image 205 is generated using a simulation method such as DEM. The virtual cross-sectional image 205 may be composed of data sampled from an image of the virtual three-dimensional structure 207. That is, the virtual cross-sectional image 205 may be generated by acquiring a cross-section of the three-dimensional structure shown by the virtual three-dimensional structure 207 as 2D data and associating it with the material properties of the material having the three-dimensional structure.

[0048] The measured cross-sectional image 206 is a data set including 2D data showing a cross-section of a material and data associated with material properties of the material having the cross-section shown by the 2D data. The measured cross-sectional image 206 is generated by measuring an experimental sample using SEM or the like. The measured cross-sectional image 206 may be composed of data sampled from an image of the measured three-dimensional structure 208. That is, the measured cross-sectional image 206 may be generated by acquiring a cross-section of the three-dimensional structure shown by the measured three-dimensional structure 208 as 2D data and associating it with the material properties of the material having the three-dimensional structure.

[0049] The virtual three-dimensional structure 207 is a data set including 3D data showing a three-dimensional structure of a material and data associated with material properties of the material having the three-dimensional structure shown by the 3D data. The virtual three-dimensional structure 207 is generated using a simulation method such as DEM.

[0050] The measured three-dimensional structure 208 is a dataset that includes 3D data showing the three-dimensional structure of a material, and data linked to the material properties possessed by the material having the three-dimensional structure shown by the 3D data. The measured three-dimensional structure 208 is generated by measuring experimental samples using a FIB-SEM or the like.

[0051] The cross-sectional image characteristic prediction model learning unit 201 learns a cross-sectional image characteristic prediction model that predicts material properties corresponding to the material structure shown in a cross-sectional image, based on at least one of the virtual cross-sectional image 205 and the measured cross-sectional image 206, and outputs it to the DB unit 110. For learning the cross-sectional image characteristic prediction model, machine learning methods such as convolutional neural networks, Vision Transformer, and ensemble learning combining these can be used. The cross-sectional image characteristic prediction model learning unit 201 may also learn the cross-sectional image characteristic prediction model using transfer learning that combines the virtual cross-sectional image 205 and the measured cross-sectional image 206.

[0052] The cross-sectional image generation model learning unit 202 learns a cross-sectional image generation model 31 based on at least one of the virtual cross-sectional image 205 and the measured cross-sectional image 206, and outputs it to the DB unit 110. The cross-sectional image generation model 31 is a generation model that generates a cross-sectional image showing the cross-section of a material based on generation conditions that include information indicating the material's properties or structure. Various known algorithms, such as GANs, VAEs, or diffusion models, can be used for the cross-sectional image generation model 31. In addition, conditional generation models such as c-GANs (conditional Generative Adversarial Networks) or conditional diffusion models may be used for the cross-sectional image generation model 31. Furthermore, a generation model such as style-GAN, which can control the geometric features of the material using style transformation, may be used for the cross-sectional image generation model 31.

[0053] Furthermore, the cross-sectional image generation model learning unit 202 may extract features using the trained cross-sectional image characteristic prediction model learned by the cross-sectional image characteristic prediction model learning unit 201, and use the extracted features when training the cross-sectional image generation model 31.

[0054] The 3D structure property prediction model learning unit 203 learns a 3D structure property prediction model that predicts material properties corresponding to the material structure shown by the 3D structure, based on at least one of the virtual 3D structure 207 and the measured 3D structure 208, and outputs it to the DB unit 110. For learning the 3D structure property prediction model, machine learning methods such as convolutional neural networks, Vision Transformer, and ensemble learning combining these can be used. The 3D structure property prediction model learning unit 203 may also learn the 3D structure property prediction model using transfer learning that combines the virtual 3D structure 207 and the measured 3D structure 208.

[0055] The 3D structure generation model learning unit 204 learns a 3D structure generation model 32 based on at least one of the virtual 3D structure 207 and the measured 3D structure 208, and outputs it to the DB unit 110. The 3D structure generation model 32 is a generation model that generates the 3D structure of a material based on a cross-sectional image. For example, a conditional generation model is used for the 3D structure generation model 32. Various known algorithms, such as c-GANs or conditional diffusion models, may be used for the conditional generation model.

[0056] Furthermore, the three-dimensional structure generation model 32 may also have a function to generate a three-dimensional structure that is easy to evaluate in the evaluation unit 129 described later. For example, the three-dimensional structure generation model 32 may use a diffusion model that generates a three-dimensional structure to which periodic boundary conditions can be applied when evaluating the three-dimensional structure. The three-dimensional structure generation model 32 is a model that has been trained so that the three-dimensional structure 23 generated by the three-dimensional structure generation model 32 satisfies predetermined geometric conditions. The predetermined geometric conditions are, for example, the condition that for a certain cross section included in the generated three-dimensional structure, there exists a cross section parallel to that cross section that has the same structure. In other words, the predetermined geometric conditions are, for example, the condition that opposing cross sections included in the three-dimensional structure have the same structure and form a periodic boundary.

[0057] Furthermore, the 3D structure generation model learning unit 204 may extract features using the trained 3D structure characteristic prediction model learned by the 3D structure characteristic prediction model learning unit 203, and use the extracted features when training the 3D structure generation model 32.

[0058] [Configuration of the Design Unit] Next, the configuration of the design unit according to the embodiment will be described. Figure 4 is a block diagram showing the functional configuration of the design unit 120 according to the embodiment. The design system 10 shown in Figure 4 comprises a DB unit 110, a design unit 120, and a UI unit 130. The design unit 120 also comprises a search unit 121, a generation unit 122, a prediction unit 125, a verification unit 128, and an evaluation unit 129. The generation unit 122 also comprises a cross-sectional image generation unit 123 and a three-dimensional structure generation unit 124. The prediction unit 125 also comprises a cross-sectional image characteristic prediction unit 126 and a three-dimensional structure characteristic prediction unit 127.

[0059] The UI unit 130 receives user input and outputs it to the DB unit 110 and the design unit 120. The user uses the UI unit 130 to input desired properties of the material to be designed, generation conditions in the cross-sectional image generation unit 123 and the three-dimensional structure generation unit 124, and verification conditions to be used in the verification unit 128. The UI unit 130 also presents the data input from the DB unit 110 and the design unit 120 to the user in text and / or graphically. The UI unit 130 presents the user with the generation conditions determined by the search unit 121, the cross-sectional image 22 generated by the cross-sectional image generation unit 123, the three-dimensional structure 23 generated by the three-dimensional structure generation unit 124, the predicted properties predicted by the cross-sectional image property prediction unit 126, the predicted properties predicted by the three-dimensional structure property prediction unit 127, and the evaluation properties evaluated by the evaluation unit 129. An example of the display screen shown on the UI unit 130 will be described later with reference to Figures 7, 8, and 9.

[0060] The DB unit 110 is implemented, for example, by a storage device or memory device such as a hard disk drive in a data management server or data center. The DB unit 110 stores experimental and simulation data related to material structure, material properties linked to the structure, and a history of previously set parameters, as will be described in more detail below. The DB unit 110 may also store information on simulation models, generation models, and prediction models that have been used in the past.

[0061] The design unit 120 performs information processing based on the setting information input from the UI unit 130 and the information acquired from the DB unit 110. The setting information includes desired characteristics, generation conditions, and verification conditions. The design unit 120 outputs the information processing process and results to the DB unit 110 for storage, or outputs them to the UI unit 130 for display to the user.

[0062] The search unit 121 determines the generation conditions 21 based on the desired characteristics input by the user, the predicted characteristics predicted by the cross-sectional image characteristics prediction unit 126 and the three-dimensional structure characteristics prediction unit 127, and a comparison thereof. The search unit 121 outputs the determined generation conditions 21 to the generation unit 122.

[0063] Here, the properties in the desired properties and predicted properties refer to the material's properties, such as its physical properties, chemical properties, or geometric features of its structure. Physical properties of a material include, for example, thermal properties such as thermal conductivity, electrical properties such as dielectric strength, and magnetic properties such as magnetic permeability. Chemical properties of a material refer to properties related to the material's reactions, such as hygroscopicity, chemical resistance, and flammability. Geometric features of a material include particle packing density and particle size distribution of the particles constituting the material. Thus, a material possesses properties of multiple characteristics.

[0064] The search unit 121 determines the generation conditions 21 using, for example, a black-box optimization method such as Bayesian optimization or a genetic algorithm, with the difference between the desired characteristics and the predicted characteristics as the evaluation function. Alternatively, the search unit 121 may determine the generation conditions 21 using, for example, reinforcement learning, with the difference between the desired characteristics and the predicted characteristics as the reward. Furthermore, the search unit 121 may use the generation conditions 21 set by the user through the UI unit 130.

[0065] The generation unit 122 generates a cross-sectional image 22 and a three-dimensional structure 23 based on the generation conditions 21 input from the search unit 121. The generation unit 122 outputs the generated cross-sectional image 22 and three-dimensional structure 23 to the UI unit 130 and the verification unit 128.

[0066] The cross-sectional image generation unit 123 generates a cross-sectional image 22 based on the generation conditions input from the search unit 121, using a trained cross-sectional image generation model 31 stored in the DB unit 110. The cross-sectional image generation unit 123 outputs the generated cross-sectional image 22 to the three-dimensional structure generation unit 124, the cross-sectional image characteristic prediction unit 126, and the verification unit 128.

[0067] The 3D structure generation unit 124 uses a learned 3D structure generation model 32 stored in the DB unit 110 to generate a 3D structure 23 based on the cross-sectional image 22 generated by the cross-sectional image generation unit 123. The 3D structure generation unit 124 outputs the generated 3D structure 23 to the 3D structure characteristic prediction unit 127 and the verification unit 128.

[0068] The prediction unit 125 uses a trained characteristic prediction model to predict the material properties that a material having the cross-section shown in the cross-sectional image 22 or the material having the three-dimensional structure 23 is expected to possess, based on the cross-sectional image 22 or the three-dimensional structure 23 input from the generation unit 122. The prediction unit 125 outputs the predicted characteristics to the search unit 121 and the verification unit 128. The trained characteristic prediction model used in the prediction unit 125 includes a trained cross-sectional image characteristic prediction model and a trained three-dimensional structure characteristic prediction model.

[0069] The cross-sectional image characteristic prediction unit 126 uses a trained cross-sectional image characteristic prediction model stored in the DB unit 110 to predict the predicted characteristic 1 based on the cross-sectional image 22 input from the cross-sectional image generation unit 123. The predicted characteristic 1 is a material characteristic corresponding to the material structure shown in the cross-sectional image 22. The cross-sectional image characteristic prediction unit 126 outputs the predicted characteristic 1 to the search unit 121 and the verification unit 128.

[0070] The 3D structure characteristic prediction unit 127 uses a learned cross-sectional image characteristic prediction model stored in the DB unit 110 to predict the predicted characteristics 2 based on the 3D structure 23 input from the 3D structure generation unit 124. The predicted characteristics 2 are material properties corresponding to the material structure shown by the 3D structure 23. The 3D structure characteristic prediction unit 127 outputs the predicted characteristics 2 to the search unit 121 and the verification unit 128.

[0071] The verification unit 128 outputs the cross-sectional image 22 and three-dimensional structure 23 input by the generation unit 122, as well as the predicted characteristics input by the prediction unit 125, to the UI unit 130. The predicted characteristics include predicted characteristic 1 and predicted characteristic 2. The verification unit 128 also outputs the three-dimensional structure 23 that satisfies the evaluation conditions to the evaluation unit 129. The evaluation conditions are, for example, whether the material properties corresponding to the material structure shown by the three-dimensional structure 23 satisfy the desired characteristics input by the user. The evaluation conditions may be set by user input via the UI unit 130.

[0072] The verification unit 128 may, for example, determine the three-dimensional structure 23 to output to the evaluation unit 129 based on evaluation conditions set by user input obtained from the UI unit 130. Alternatively, the verification unit 128 may determine the three-dimensional structure 23 to output to the evaluation unit 129 using an algorithm based on the difference or ratio between predicted characteristics and desired characteristics.

[0073] The evaluation unit 129 evaluates the material properties of the three-dimensional structure 23 input from the verification unit 128 and outputs them to the UI unit 130 as realized properties. While predicted properties are properties predicted using machine learning methods based on the relationship between past data properties and three-dimensional structures, realized properties are the properties that the material corresponding to the material structure shown by the three-dimensional structure 23 actually possesses. The properties in realized properties have the same meaning as the properties in desired properties and predicted properties described above. The evaluation unit 129 evaluates the realized properties using, for example, a three-dimensional physical simulation such as FEM (Finite Element Method). Alternatively, the evaluation unit 129 may evaluate the realized properties by communicating with an external experimental device and conducting experiments.

[0074] [Operation] Next, the operation of the design unit 120 in the embodiment of this disclosure will be described with reference to Figure 5. Figure 5 is a flowchart illustrating the operation of the design unit 120 according to the embodiment.

[0075] First, the search unit 121 acquires the desired characteristics (S11). Specifically, the search unit 121 receives input of the desired characteristics from the user using the UI unit 130 and acquires the desired characteristics. The user sets the desired characteristics of the material to be designed via the UI unit 130. For example, the user sets the desired characteristics of the material by operating the touch panel and inputting the type and value of the material characteristics. The desired characteristics may be input by specifying the value of the material characteristics, or by specifying a range of values ​​for the material characteristics.

[0076] Next, the search unit 121, the generation unit 122, and the prediction unit 125 sequentially generate cross-sectional images 22 and three-dimensional structures 23 of the material that realize the desired characteristics (S12). The processes performed in step S12 will be described in detail later with reference to Figure 6.

[0077] Once the cross-sectional image 22 and the three-dimensional structure 23 are generated, the verification unit 128 verifies the three-dimensional structure 23 based on the corresponding cross-sectional image 22 and predicted characteristics (S13). Verifying the three-dimensional structure 23 means determining whether or not the three-dimensional structure 23 satisfies the evaluation conditions. This determination is made, for example, by accepting a user's judgment. Specifically, the verification unit 128 presents this information to the user by displaying the three-dimensional structure 23, as well as the corresponding cross-sectional image 22 and predicted characteristics, on the UI unit 130. Based on this presented information, the user selects a three-dimensional structure 23 from among the presented three-dimensional structures 23 whose realization characteristics are to be evaluated via the UI unit 130.

[0078] The verification unit 128 outputs the three-dimensional structure 23 selected by the user to the evaluation unit 129.

[0079] Examples of the display screens presented to the user during the processing in step S13 will be described later using Figures 7 and 8.

[0080] Finally, the evaluation unit 129 evaluates the realization characteristics of the three-dimensional structure 23 (S14). Specifically, the evaluation unit 129 evaluates the material properties of the material corresponding to the material structure shown by the three-dimensional structure 23. The evaluation unit 129 presents these material properties as realization characteristics to the user using the UI unit 130.

[0081] Next, the process performed in step S12 of Figure 5 will be explained with reference to Figure 6. Figure 6 is a flowchart illustrating the operation of the design unit in S12 of Figure 5.

[0082] First, the search unit 121 determines the generation conditions 21 (S21). Specifically, in step S11 of Figure 5, the search unit 121 determines the generation conditions 21 based on the desired characteristics input by the user. In this way, the search unit 121 sets the generation conditions 21. The search unit 121 outputs the set generation conditions 21 to the generation unit 122.

[0083] Next, the cross-sectional image generation unit 123 generates a cross-sectional image 22 based on the acquired generation conditions 21 (S22). Specifically, the cross-sectional image generation unit 123 generates a cross-sectional image 22 based on the generation conditions 21 using the cross-sectional image generation model 31 stored in the DB unit 110. The cross-sectional image generation unit 123 outputs the generated cross-sectional image 22 to the cross-sectional image characteristic prediction unit 126.

[0084] The cross-sectional image characteristic prediction unit 126 predicts material properties based on the input cross-sectional image 22 (S23). Specifically, the cross-sectional image characteristic prediction unit 126 uses a trained cross-sectional image characteristic prediction model stored in the DB unit 110 to predict the predicted characteristics 1 based on the cross-sectional image 22 input from the cross-sectional image generation unit 123. The cross-sectional image characteristic prediction unit 126 outputs the predicted characteristics 1 to the search unit 121.

[0085] Next, the search unit 121 performs a determination of whether or not to change the generation conditions 21 (S24). This determination is performed, for example, based on whether or not the predicted characteristics 1 output by the cross-sectional image characteristic prediction unit 126 satisfy the desired characteristics specified by the user. In other words, if the search unit 121 determines that the predicted characteristics 1 do not satisfy the desired characteristics, it decides to change the generation conditions 21. In this way, the search unit 121 determines whether or not the predicted characteristics 1 satisfy a predetermined condition. The predetermined condition is that the predicted characteristics 1 satisfy the desired characteristics.

[0086] If the cross-sectional image generation model 31 is a probabilistic model rather than a deterministic model, the determination may be made, for example, based on whether or not the number of cross-sectional images 22 set in advance by the user has been reached. In other words, the search unit 121 may decide to change the generation condition 21 if it determines that the number of cross-sectional images 22 generated based on the generation condition 21 is equal to or greater than a predetermined number. The predetermined number is, for example, 10, but may also be 20. Here, the predetermined number is not particularly limited. The predetermined number may be set in advance by user input via the UI unit 130.

[0087] If it is determined that the generation conditions should be changed (Yes in S24), in other words, if it is determined that the predicted characteristic 1 does not satisfy the desired characteristics, the process returns to step S21, and the search unit 121 determines the generation conditions 21. At this time, the search unit 121 may determine the generation conditions 21 based on the desired characteristics and the predicted characteristic 1. In this way, the search unit 121 adjusts the generation conditions 21 based on the predicted characteristics 1, etc.

[0088] If it is determined that the generation condition 21 does not need to be changed (No in S24), in other words, if it is determined that the predicted characteristic 1 satisfies the desired characteristic, the search unit 121 presents the cross-sectional image 22 and the corresponding predicted characteristic 1 to the user by displaying them on the display unit of the UI unit 130 (S25). The search unit 121 can also present the information to the user by displaying a display screen, which will be described later, as shown in Figure 7 or Figure 8.

[0089] Next, the search unit 121 performs a determination of whether or not to change the generation conditions 21 (S26). Specifically, the search unit 121 obtains an instruction from the user indicating whether or not to change the generation conditions, and performs a determination of whether or not to change the generation conditions 21 based on that instruction. In the process of step S25, the user verifies the cross-sectional image 22 and the corresponding prediction characteristics 1 displayed on the display unit of the UI unit 130 based on the domain knowledge possessed by the user. In other words, the user judges, in light of their own knowledge, whether or not the cross-section shown in the cross-sectional image 22 is a possible cross-section for the material to be designed, and inputs the result of this judgment using the UI unit 130. If the user determines that the cross-section shown in the cross-sectional image 22 is an impossible cross-section for the material to be designed, they input an instruction indicating that they want to change the generation conditions 21.

[0090] If it is determined that the generation conditions should be changed (Yes in S26), in other words, if the user has given an instruction to change the generation conditions 21, the process returns to step S21, and the search unit 121 determines the generation conditions 21. At this time, the search unit 121 may determine the generation conditions 21 based on information including the desired characteristics, the predicted characteristics 1, and the user's judgment. Information including the user's judgment is, for example, information indicating an operation performed by the user on the cross-sectional image 22 displayed on the UI unit 130. This operation is, for example, a weighted synthesis operation of two or more cross-sectional images 22, but it may also be an operation to change the value of the generation conditions 21 displayed in association with the cross-sectional image 22.

[0091] If it is determined that the generation conditions will not be changed (No in S26), in other words, if the user has given an instruction not to change the generation conditions 21, the 3D structure generation unit 124 generates a 3D structure 23 based on the cross-sectional image 22 (S27). Specifically, the 3D structure generation unit 124 uses the learned 3D structure generation model 32 stored in the DB unit 110 to generate the 3D structure 23 based on the cross-sectional image 22 generated by the cross-sectional image generation unit 123. The 3D structure generation unit 124 outputs the generated 3D structure 23 to the 3D structure characteristic prediction unit 127.

[0092] The three-dimensional structure property prediction unit 127 predicts material properties based on the three-dimensional structure 23 (S28). Specifically, the three-dimensional structure property prediction unit 127 uses a trained cross-sectional image property prediction model stored in the DB unit 110 to predict predicted properties 2 based on the three-dimensional structure 23 input from the three-dimensional structure generation unit 124. The three-dimensional structure property prediction unit 127 outputs the predicted properties 2 to the search unit 121.

[0093] Next, the search unit 121 determines whether or not to change the generation conditions 21 (S29). This determination is made, for example, based on whether or not the predicted characteristics 2 output by the three-dimensional structure characteristic prediction unit 127 satisfy the desired characteristics specified by the user. In other words, if the search unit 121 determines that the predicted characteristics 2 do not satisfy the desired characteristics, it decides to change the generation conditions 21.

[0094] Alternatively, the determination may be performed based, for example, on whether the number of three-dimensional structures 23 set in advance by the user has been reached. In other words, the search unit 121 may decide to change the generation condition 21 if it determines that the number of three-dimensional structures 23 generated based on the cross-sectional image 22 is equal to or greater than a predetermined number. The predetermined number is, for example, 5, but may also be 10. Here, the predetermined number is not particularly limited. The predetermined number may be set in advance by user input via the UI unit 130.

[0095] If it is determined that the generation condition 21 should be changed (Yes in S29), the process returns to step S21, and the search unit 121 determines the generation condition 21. At this time, the search unit 121 may determine the generation condition 21 based on information including the desired characteristics, predicted characteristics 1, predicted characteristics 2, and the user's judgment. In other words, the search unit 121 adjusts the generation condition based on the predicted characteristics 2, etc.

[0096] If it is determined that the generation condition 21 is not to be changed (No in S29), the process proceeds to step S13 in Figure 5. In other words, the generation unit 122 outputs the generated cross-sectional image 22 and three-dimensional structure 23 to the verification unit 128, and the prediction unit 125 outputs the predicted characteristics to the verification unit 128. The verification unit 128 verifies the input three-dimensional structure 23 based on the corresponding cross-sectional image 22 and predicted characteristics.

[0097] According to the operation described above using Figures 5 and 6, the design unit 120 generates a cross-sectional image 22 based on the generation conditions 21, and generates a three-dimensional structure 23 based on the evaluation of the generated cross-sectional image 22. In other words, the design system 10 according to the embodiment can efficiently design materials by using a cross-sectional image as an intermediate representation of the material structure and generating and evaluating the cross-sectional image and three-dimensional structure in stages. Furthermore, by performing the design in this way, materials that do not have a periodic structure or a series structure, such as composite materials, can be efficiently designed.

[0098] [Example of display screen] Below, an example of a display screen shown on the UI unit 130 in the embodiment will be described with reference to Figures 7 to 9.

[0099] First, an example of a display image used when the cross-sectional structure, three-dimensional structure, and predictive characteristics are displayed on the display screen of the UI unit 130 in the embodiment will be described with reference to Figure 7. Figure 7 is a diagram showing an example of a display screen displayed on the UI unit 130 according to the embodiment.

[0100] Figure 7 shows a display screen that shows the ID (identification), generation conditions, cross-sectional image, three-dimensional structure, predicted characteristics, and characterization in a tabular format. The predicted characteristics shown in Figure 7 are predicted characteristics A. Predicted characteristics A are predicted characteristics related to one of several characteristics possessed by the material.

[0101] ID 25 indicates the ID of the dataset output in the design system 10. ID 25 may be set to distinguish the generation conditions 21 determined by the search unit 121. ID 25 may also be set to distinguish the cross-sectional image 22 or the three-dimensional structure 23 generated by the generation unit 122. ID 25 may also be set to distinguish the three-dimensional structure 23 when the verification unit 128 presents the three-dimensional structure 23 to the user. The generation conditions 21, the cross-sectional image 22, the three-dimensional structure 23, and the prediction characteristics 26 are associated with ID 25.

[0102] The generation condition 21 represents the generation condition associated with ID 25. The first generation condition 211 represents one element of the vector of generation condition 21. In other words, the first generation condition 211 is a condition that corresponds to one piece of information among the information indicating the properties or structure of multiple materials included in generation condition 21. The first generation condition 211 may be displayed together with a slider bar, for example. In Figure 7, the lower limit 212 of the first generation condition, the upper limit 213 of the first generation condition, and the slider bar indicating the value 214 of the first generation condition are displayed associated with the first generation condition 211.

[0103] Cross-sectional image 22 shows the cross-sectional image corresponding to ID 25.

[0104] The three-dimensional structure 23 shows the three-dimensional structure corresponding to ID 25.

[0105] Prediction characteristic 26 indicates one of the prediction characteristics corresponding to ID 25. Prediction characteristic 261 indicates prediction characteristic 1 output by the cross-sectional image characteristic prediction unit 126, and prediction characteristic 262 indicates prediction characteristic 2 output by the three-dimensional structure characteristic prediction unit 127. Desired characteristic 263 indicates the desired characteristic entered by the user.

[0106] Checkbox 27 indicates a checkbox that accepts the user's judgment as to whether or not to evaluate the three-dimensional structure 23 corresponding to ID 25 in the evaluation unit 129 during the process of step S13 in Figure 5. For example, in the process of step S13 in Figure 5, the user inputs an instruction not to change the generation conditions 21 by checking the checkbox 27 corresponding to ID 25 of the three-dimensional structure 23 that is determined to satisfy the evaluation conditions.

[0107] The display screen shown in Figure 7 is shown on the UI unit 130 during processes such as step S13 in Figure 5 and step S25 in Figure 6. However, at this time, information that has not yet been generated when the process is executed is not displayed.

[0108] In the process of transitioning from step S26 to step S21 in Figure 6, the information including the user's judgment may, for example, be information indicating an operation performed by the user on the display screen shown in Figure 7. Such operation may, for example, be an operation to change the value 214 of the generation condition displayed on the slider bar by dragging. At this time, the search unit 121 determines the value 214 of the generation condition that was changed by the user's operation as the next generation condition 21. Alternatively, such operation may, for example, be an operation of not checking the checkbox 27. At this time, the search unit 121 lowers the weight of the generation condition 21 associated with the checkbox 27 that was not checked and determines the next generation condition 21.

[0109] Furthermore, in the embodiments of this disclosure, another example of a display image used when the cross-sectional structure, three-dimensional structure, and predictive characteristics are displayed on the display screen of the UI unit 130 will be described with reference to Figure 8. Figure 8 is a diagram showing another example of a display screen displayed on the UI unit 130 according to the embodiment.

[0110] In the graph shown in Figure 8, the vertical axis represents the value of the first predicted characteristic 301, and the horizontal axis represents the value of the second predicted characteristic 302. The first predicted characteristic 301 is, for example, predicted characteristic A, and the second predicted characteristic 302 is, for example, predicted characteristic B. Predicted characteristic A and predicted characteristic B each represent a predicted characteristic for one of several characteristics of the material. The graph consisting of the first predicted characteristic 301 and the second predicted characteristic 302 may be a two-dimensional scatter plot. Furthermore, it may also be a three-dimensional scatter plot combined with other predicted characteristics.

[0111] In the graph shown in Figure 8, the white circles represent prediction characteristic 1 associated with the cross-sectional image 22, and the black circles represent prediction characteristic 2 associated with the three-dimensional structure 23. The dashed lines connecting the white and black circles indicate the correspondence between the white and black circles. In other words, in the graph shown in Figure 8, for each pair of cross-sectional image 22 and three-dimensional structure 23 associated with the same ID 25, the prediction characteristic 303 of the cross-sectional image, the prediction characteristic 304 of the three-dimensional structure, and the difference 305 of the prediction characteristics are shown.

[0112] The display screen shown in Figure 8 is displayed on the UI unit 130, for example, during the process in step S13 of Figure 5, similar to the display screen shown in Figure 7.

[0113] Next, an example of a display screen shown on the UI unit 130 when the generation conditions are changed in the embodiments of this disclosure will be described with reference to Figure 9. Figure 9 is a diagram showing yet another example of a display screen shown on the UI unit 130 according to the embodiment. Using the display screen shown in Figure 9, the step in which the search unit 121 determines the value of the generation conditions, which is performed when the generation conditions are changed after the processing of step S26 in Figure 6, will also be described.

[0114] The first generation condition 211 represents one element of the vector of generation conditions 21 generated by the search unit 121. In the display screen shown in Figure 9, the first generation condition 211 is displayed with a slider bar corresponding to the first generation condition 211, which represents the lower limit 212 of the first generation condition, the upper limit 213 of the first generation condition, and the value 214 of the first generation condition. In the process of step S26 in Figure 6, the value 214 of the first generation condition can be used when generating a new generation condition 21.

[0115] Furthermore, the rectangular area labeled "Confirm" in Figure 9 is the generate button 401. The user can input an instruction to change the generation conditions 21 by selecting the generate button 401 via the UI unit 130. In other words, by the user selecting the generate button 401, the process moves from step S26 to step S21.

[0116] Furthermore, the rectangular area labeled "+0.2" in Figure 9 is the weight coefficient input box 402, and the roughly rectangular area enclosed by the dashed line surrounding the cross-sectional image 22 is the cross-sectional image input box 403.

[0117] In step S26 of Figure 6, the user can change the first generation condition 211 to any value by sliding the value 214 of the first generation condition on the slider bar. At this time, a new cross-sectional image 22 is generated based on the generation condition 21, which includes the first generation condition 211 that has been changed to the arbitrary value. Subsequently, the process proceeds, and in step S25 of Figure 6, the user can check the cross-sectional image and prediction characteristics on a display screen, for example, as shown in Figure 8.

[0118] Furthermore, in step S26 of Figure 6, the user can set a new generation condition 21 by weighting and combining two or more cross-sectional images 22, rather than directly inputting the value 214 of the first generation condition. The user inputs multiple cross-sectional images 22, for example, displayed on the display screen shown in Figure 7, into the cross-sectional image input box 403 by dragging and dropping, and inputs the weight coefficient corresponding to each cross-sectional image 22 into the weight coefficient input box 402. This allows the user to execute and set the weighted combination of cross-sectional images 22.

[0119] In the display screen shown in Figure 9, three sets of weight coefficient input boxes 402 and cross-sectional image input boxes 403 are shown, and weighted synthesis of the cross-sectional image 22 is performed using two of these sets. Specifically, in the display screen shown in Figure 9, the weight of the cross-sectional image 22 input into the left cross-sectional image input box 403 is set to "+0.2", and the weight of the cross-sectional image 22 input into the middle cross-sectional image input box 403 is set to "+0.8", and weighted synthesis of the cross-sectional image 22 is performed.

[0120] The generation conditions 21 for weighted synthesis of cross-sectional images 22 are determined, for example, using the following formula.

[0121]

[0122] Here, z is a vector representing the new generation condition 21. The value of the first generation condition 211 included in the generation condition 21 corresponds to the value 214 of the first generation condition. In other words, the user can adjust the value of the weight coefficient while checking the slider bar. Also, c is a weight coefficient, and its value corresponds to the value entered in the weight coefficient input box 402. x is a vector representing the prediction characteristics associated with the cross-sectional image 22. g is a function that converts the cross-sectional image 22 to the corresponding generation condition 21. For example, if the cross-sectional image generation unit 123 uses GANs as the cross-sectional image generation model 31, a known method related to GAN Inversion is used for g.

[0123] (Summary) As explained above, according to one aspect of this disclosure, by using a cross-sectional image as an intermediate representation of the material structure and generating and evaluating the cross-sectional image and three-dimensional structure in stages, it is possible to efficiently design materials that do not have a periodic structure or a series structure, such as composite materials.

[0124] Furthermore, when designing materials on an experimental basis, it is necessary to actually create the materials. Since the manufacturing methods or materials themselves of unknown materials may be dangerous, according to one aspect of this disclosure, materials can be designed more safely.

[0125] Furthermore, while AI (Artificial Intelligence)-based design methods generally suffer from the problem of low interpretability and credibility of the generated three-dimensional structures, according to one aspect of this disclosure, after generating a cross-sectional image, evaluation based on the user's expertise is accepted before generating the three-dimensional structure, thereby improving the interpretability and credibility of the generated three-dimensional structure.

[0126] (Other Embodiments) Although various embodiments have been described above with reference to the drawings, it goes without saying that this disclosure is not limited to these examples. It will be obvious to those skilled in the art that various modifications or alterations can be conceived within the scope of the claims, and these will naturally also fall within the technical scope of this disclosure. Furthermore, the components of the above embodiments may be combined in any way without departing from the spirit of the disclosure.

[0127] In the embodiments described above, this disclosure has described examples in which the disclosure is configured using hardware, but this disclosure can also be implemented in software in conjunction with hardware.

[0128] Furthermore, each functional block used in the description of the above embodiments is typically implemented as an integrated circuit, or LSI (Large Scale Integration). The integrated circuit controls each functional block used in the description of the above embodiments and may have inputs and outputs. These may be individually integrated on a single chip, or some or all of them may be integrated on a single chip. Here, we refer to it as an LSI, but depending on the degree of integration, it may also be called an IC (Integrated Circuit), system LSI, super LSI, or ultra LSI.

[0129] Furthermore, the method of integrated circuit implementation is not limited to LSIs; it may also be realized using dedicated circuits or general-purpose processors. After LSI manufacturing, FPGAs (Field Programmable Gate Arrays) that can be programmed, or reconfigurable processors that can reconfigure the connections or settings of circuit cells inside the LSI, may also be used.

[0130] Furthermore, if advances in semiconductor technology or other derived technologies lead to the emergence of integrated circuit technologies that can replace LSIs, then naturally, functional blocks can be integrated using those technologies. Applications of biotechnology and optical integrated circuits are among the possibilities.

[0131] Furthermore, the general or specific embodiments of this disclosure may be implemented as a system, apparatus, method, integrated circuit, or computer program. Alternatively, they may be implemented as a computer-readable non-temporary recording medium such as an optical disk, HDD, or semiconductor memory on which the computer program is stored. They may also be implemented as any combination of a system, apparatus, method, integrated circuit, computer program, and recording medium.

[0132] Furthermore, each of the above embodiments may be modified, replaced, added, or omitted in various ways within the scope of the claims or their equivalents.

[0133] <Summary of this Disclosure> The aspects of this disclosure derived from the above configuration, operation, and GUI are, for example, the following. Below, the aspects of this disclosure derived from the content of this specification will be explained together with the effects obtained by such aspects of this disclosure.

[0134] A material design apparatus according to a first aspect of this disclosure includes: a search unit that sets generation conditions for generating a three-dimensional structure of a material, including information indicating at least one of the material's properties and structure; a cross-sectional image generation unit that generates a cross-sectional image of the material based on the generation conditions set by the search unit; and a three-dimensional structure generation unit that generates a three-dimensional structure of the material based on the cross-sectional image generated by the cross-sectional image generation unit. The design system 10 in the above embodiment is an example of a material design apparatus.

[0135] Such a material design apparatus can design materials more efficiently by using a cross-sectional image 22 as an intermediate representation of the material structure and generating the cross-sectional image 22 and the three-dimensional structure 23 in a stepwise manner. As shown in Figure 2, such a material design apparatus can generate the three-dimensional structure 23 stably and at low cost by generating the cross-sectional image 22 and the three-dimensional structure 23 in a stepwise manner using a cross-sectional image generation model 31 and a three-dimensional structure generation model 32. Furthermore, such a material design apparatus can generate the three-dimensional structure of any material, regardless of whether the material has a periodic structure or a sequential structure.

[0136] A material design apparatus according to a second aspect of this disclosure is a material design apparatus according to a first aspect, wherein the material is a material that does not have a periodic structure or a series structure.

[0137] Such material design equipment allows users to safely and cost-effectively perform structural design using generative models, even for materials that do not have a periodic or sequential structure, by using two-dimensional cross-sectional images as an intermediate representation of the material structure.

[0138] A material design apparatus according to a third aspect of the present disclosure is a material design apparatus according to the first or second aspect, further comprising a cross-sectional image characteristic prediction unit that predicts the properties of a material having the cross-section shown in the cross-sectional image from the cross-sectional image generated by the cross-sectional image generation unit, and the search unit performs adjustment of the generation conditions based on the properties predicted by the cross-sectional image characteristic prediction unit.

[0139] Such a material design apparatus can also improve the accuracy of the generated three-dimensional structure 23 in relation to the desired characteristics by performing an evaluation of the cross-sectional image 22 generated by the cross-sectional image generation model 31 based on predictive characteristics and validity, thereby contributing to the reduction of design costs.

[0140] A material design apparatus according to a fourth aspect of the present disclosure is a material design apparatus according to any one of the first to third aspects, further comprising a three-dimensional structure property prediction unit that predicts the properties of a material having the three-dimensional structure from the three-dimensional structure generated by the three-dimensional structure generation unit, and the search unit adjusts the generation conditions based on the properties predicted by the three-dimensional structure property prediction unit.

[0141] Such a material design apparatus can improve the accuracy of the generated three-dimensional structure 23 in relation to the desired properties by performing an evaluation of the three-dimensional structure 23 generated by the three-dimensional structure generation model 32 based on its predicted properties and validity, thereby contributing to the reduction of design costs.

[0142] A material design apparatus according to a fifth aspect of the present disclosure is a material design apparatus according to any one of the first to fourth aspects, wherein the three-dimensional structure generation unit uses a probabilistic model that can generate different three-dimensional structures when the same cross-sectional image is input as a three-dimensional structure generation model for generating the three-dimensional structure from the cross-sectional image.

[0143] Such a material design device uses a probabilistic model for the three-dimensional structure generation model 32, which allows the learning of the three-dimensional structure generation model 32 in the three-dimensional structure generation model learning unit to be performed stably using methods such as maximum likelihood estimation, thus enabling the generation of more stable three-dimensional structures of materials.

[0144] A material design apparatus according to the sixth aspect of this disclosure is a material design apparatus according to any one of the first to fifth aspects, wherein the cross-sectional image generation unit uses a definitive model that generates the same cross-sectional image when the same generation conditions are input as a cross-sectional image generation model that generates the cross-sectional image from the generation conditions.

[0145] Such a material design apparatus uses a deterministic model for the cross-sectional image generation model 31, which reduces the variance of the objective function and reward function in the cross-sectional image search performed by the search unit 121, the cross-sectional image generation unit 123, and the cross-sectional image characteristic prediction unit 126. This allows for more efficient searching of cross-sectional images, and thus more efficient generation of the three-dimensional structure of the material. Even when a probabilistic model with large variance of characteristics is used for the three-dimensional structure generation model, such a material design apparatus can still search for cross-sectional images more efficiently, enabling the design unit 120 to efficiently search for the three-dimensional structure.

[0146] A material design apparatus according to the seventh aspect of the present disclosure is a material design apparatus according to any one of the first to sixth aspects, further comprising a learning unit that learns a cross-sectional image generation model that generates the cross-sectional image from the generation conditions and a three-dimensional structure generation model that generates the three-dimensional structure from the cross-sectional image, wherein the learning unit learns the three-dimensional structure generation model such that the three-dimensional structure generated by the three-dimensional structure generation model satisfies geometric conditions.

[0147] In this material design apparatus, when generating the three-dimensional structure 23, the cross-sectional image generation model 31 and the three-dimensional structure generation model 32 are used separately and in stages. This stabilizes the learning of the three-dimensional structure generation model 32 by the three-dimensional structure generation model learning unit 204, making it easier to impose geometric constraints on the three-dimensional structure 23 generated by the three-dimensional structure generation model 32, such as by providing periodic boundary conditions during the learning process. This improves the efficiency of the design unit 120 when searching for the three-dimensional structure 23 using the learned three-dimensional structure generation model 32.

[0148] A material design apparatus according to the eighth aspect of this disclosure is a material design apparatus according to the seventh aspect, wherein the learning unit learns the cross-sectional image generation model using measured cross-sectional images of the material and learns the three-dimensional structure generation model using the three-dimensional structure of the material generated by a simulation method.

[0149] Such a material design device can learn a cross-sectional image generation model 31 from measured cross-sectional images 206 and a three-dimensional structure generation model 32 from a virtual three-dimensional structure 207. Therefore, even in situations where measured three-dimensional structures 208, which have high measurement costs, cannot be used, the results of measured data can be reflected in the search for three-dimensional structures 23 by the design unit 120, thereby improving the accuracy of the generated three-dimensional structure 23.

[0150] A material design apparatus according to the ninth aspect of this disclosure is a material design apparatus according to the seventh aspect, wherein the learning unit learns the cross-sectional image generation model and the three-dimensional structure generation model based on a learned characteristic prediction model for predicting the properties of the material from the cross-sectional image or the three-dimensional structure. The cross-sectional image characteristic prediction model and the three-dimensional structure characteristic prediction model in the above embodiment are examples of characteristic prediction models.

[0151] In such a material design device, the learning unit 200 first completes the training of the cross-sectional image characteristic prediction model and the three-dimensional structure characteristic prediction model, and then uses the trained characteristic prediction models to train the cross-sectional image generation model 31 and the three-dimensional structure generation model 32, thereby enabling the training of a cross-sectional image generation model 31 and a three-dimensional structure generation model 32 that capture the structural characteristics of materials that are sensitive to material properties.

[0152] A material design apparatus according to the tenth aspect of this disclosure is a material design apparatus according to any one of the first to ninth aspects, further comprising an output unit that displays the cross-sectional image generated by the cross-sectional image generation unit on a display unit, and an acquisition unit that acquires a user's evaluation of the cross-sectional image output by the output unit. The search unit 121 in the above embodiment is an example of the output unit and the acquisition unit.

[0153] Such a material design apparatus, by using a GUI and operation steps that display and verify the cross-sectional image 22 to the user in steps S25 and S26 of Figure 6, allows the user to verify, using their domain knowledge, whether the generated model is generating a structure that is physically or manufacturingly impossible before generating the three-dimensional structure. This suppresses the decline in interpretability and acceptance of results that is often a problem with AI-based design methods.

[0154] A material design apparatus according to an eleventh aspect of the present disclosure is a material design apparatus according to a tenth aspect, further comprising a determination unit that determines whether the characteristics predicted to be possessed by the material in the cross-section shown in the cross-sectional image satisfy predetermined conditions, and the output unit causes the cross-sectional image whose characteristics are determined by the determination unit to satisfy predetermined conditions to be displayed on the display unit. The search unit 121 in the above embodiment is an example of a determination unit.

[0155] Such a material design apparatus performs screening using predicted characteristics by the cross-sectional image characteristic prediction unit 126 and the search unit 121 before the user evaluation in step S26 of Figure 6. This eliminates the need for the user to verify all generated cross-sectional images 22, thereby reducing the cost required for manual validation.

[0156] A material design apparatus according to a twelfth aspect of the present disclosure is a material design apparatus according to any one of the first to eleventh aspects, further comprising a display unit that displays at least one of the generation conditions, the cross-sectional image, the three-dimensional structure, the characteristics expected to be possessed by the material of the cross-section shown in the cross-sectional image, and the characteristics expected to be possessed by the material of the three-dimensional structure.

[0157] Such a material design apparatus can visualize the difference between the predicted characteristics and generation conditions for the cross-sectional image 22 and the predicted characteristics and generation conditions for the three-dimensional structure 23 on the display screen shown in Figure 7. Therefore, the user can easily determine whether there are any abnormalities in the generation of the three-dimensional structure by the three-dimensional structure generation unit 124. Consequently, if the user determines that there are abnormalities in the generation of the three-dimensional structure 23, the evaluation unit 129 will not perform the evaluation, thus avoiding the cost of evaluating the abnormal three-dimensional structure 23 by the evaluation unit 129.

[0158] A material design apparatus according to a thirteenth aspect of this disclosure is a material design apparatus according to a twelfth aspect, wherein the display unit displays the predicted properties of the material in the cross-section shown in the cross-sectional image and the predicted properties of the material in the three-dimensional structure in a manner that allows for comparison.

[0159] Such a material design apparatus visualizes the difference between the predicted characteristic values ​​for the cross-sectional image and the predicted characteristic values ​​for the three-dimensional structure on the display screen shown in Figure 8, allowing the user to easily understand whether there are any abnormalities in the generation of the three-dimensional structure by the three-dimensional structure generation unit 124. Therefore, if the user identifies a three-dimensional structure 23 as having an abnormality in its generation, the evaluation unit 129 will not perform the evaluation, thus avoiding the cost of evaluating the abnormal three-dimensional structure with the evaluation unit 129.

[0160] A material design apparatus according to a 14th aspect of the present disclosure is a material design apparatus according to a 12th or 13th aspect, further comprising an operation receiving unit that receives an operation from a user to weight-combine at least two of the cross-sectional images displayed on the display unit, and the search unit sets the generation conditions based on the operation.

[0161] With such a materials design device, as shown in the display screen in Figure 9, even when variables that have no physical meaning and are difficult to interpret directly, such as latent variables of GANs, are used in the generation conditions, the user can intuitively change the generation conditions.

[0162] A material design method according to a 15th aspect of the present disclosure is a material design method performed by a computer, comprising: a search step of setting generation conditions for generating a three-dimensional structure of a material including information indicating at least one of the properties and structure of the material; a cross-sectional image generation step of generating a cross-sectional image of the material based on the generation conditions set by the search step; and a three-dimensional structure generation step of generating a three-dimensional structure of the material based on the cross-sectional image generated by the cross-sectional image generation step.

[0163] This material design method allows for more efficient material design by using a cross-sectional image 22 as an intermediate representation of the material structure and generating the cross-sectional image 22 and the three-dimensional structure 23 in a stepwise manner.

[0164] The program relating to the sixteenth aspect of this disclosure is a program for causing a computer to execute the material design method relating to the fifteenth aspect.

[0165] Such programs can help design materials more efficiently.

[0166] This disclosure can be used in a materials design system for exploring the three-dimensional structure of novel materials.

[0167] 10 Design System 11 Storage Device 12 Processing Device 13 UI Device 14 Communication Device 15 Network 21 Generation Conditions 22 Cross-sectional Image 23, 24 Three-dimensional Structure 25 ID 26 Prediction Characteristics 27 Checkbox 31 Cross-sectional Image Generation Model 32 Three-dimensional Structure Generation Model 110 DB Unit 120 Design Unit 121 Search Unit 122 Generation Unit 123 Cross-sectional Image Generation Unit 124 Three-dimensional Structure Generation Unit 125 Prediction Unit 126 Cross-sectional Image Characteristics Prediction Unit 127 Three-dimensional Structure Characteristics Prediction Unit 128 Verification Unit 129 Evaluation Unit 130 UI Unit 200 Learning Unit 201 Cross-sectional Image Characteristics Prediction Model Learning Unit 202 Cross-sectional Image Generation Model Learning Unit 203 Three-dimensional Structure Characteristics Prediction Model Learning Unit 204 Three-dimensional Structure Generation Model Learning Unit 205 Virtual Cross-sectional Image 206 Measured Cross-Sectional Image 207 Virtual Three-Dimensional Structure 208 Measured Three-Dimensional Structure 211 First Generation Condition 212 Lower Limit of First Generation Condition 213 Upper Limit of First Generation Condition 214 Value of First Generation Condition 261 Predicted Characteristic 1 262 Predicted Characteristic 2 263 Desired Characteristic 301 First Predicted Characteristic 302 Second Predicted Characteristic 303 Predicted Characteristic of Cross-Sectional Image 304 Predicted Characteristic of Three-Dimensional Structure 305 Difference in Predicted Characteristic 401 Generate Button 402 Weight Coefficient Input Box 403 Cross-Sectional Image Input Box

Claims

1. A material design apparatus comprising: a search unit that sets generation conditions for generating a three-dimensional structure of a material, including information indicating at least one of the material's properties and structure; a cross-sectional image generation unit that generates a cross-sectional image of the material based on the generation conditions set by the search unit; and a three-dimensional structure generation unit that generates a three-dimensional structure of the material based on the cross-sectional image generated by the cross-sectional image generation unit.

2. The material design apparatus according to claim 1, wherein the material is a material that does not have a periodic structure or a series structure.

3. The material design apparatus according to claim 1 or 2, further comprising a cross-sectional image characteristic prediction unit that predicts the properties of a material having the cross-section shown in the cross-sectional image from the cross-sectional image generated by the cross-sectional image generation unit, wherein the search unit adjusts the generation conditions based on the properties predicted by the cross-sectional image characteristic prediction unit.

4. The material design apparatus according to claim 1 or 2, further comprising a three-dimensional structure property prediction unit that predicts the properties of a material having the three-dimensional structure from the three-dimensional structure generated by the three-dimensional structure generation unit, wherein the search unit adjusts the generation conditions based on the properties predicted by the three-dimensional structure property prediction unit.

5. The material design apparatus according to claim 1 or 2, wherein the three-dimensional structure generation unit uses a probabilistic model capable of generating different three-dimensional structures when the same cross-sectional image is input, as a three-dimensional structure generation model for generating the three-dimensional structure from the cross-sectional image.

6. The material design apparatus according to claim 1 or 2, wherein the cross-sectional image generation unit uses a definitive model that generates the same cross-sectional image when the same generation conditions are input, as a cross-sectional image generation model that generates the cross-sectional image from the generation conditions.

7. The material design apparatus according to claim 1 or 2, further comprising a learning unit for learning a cross-sectional image generation model for generating the cross-sectional image from the generation conditions and a three-dimensional structure generation model for generating the three-dimensional structure from the cross-sectional image, wherein the learning unit learns the three-dimensional structure generation model so that the three-dimensional structure generated by the three-dimensional structure generation model satisfies geometric conditions.

8. The material design apparatus according to claim 7, wherein the learning unit learns the cross-sectional image generation model using cross-sectional images of the material measured, and learns the three-dimensional structure generation model using the three-dimensional structure of the material generated by a simulation method.

9. The material design apparatus according to claim 7, wherein the learning unit learns the cross-sectional image generation model and the three-dimensional structure generation model based on a learned property prediction model for predicting the properties of the material from the cross-sectional image or the three-dimensional structure.

10. The material design apparatus according to claim 1 or 2, further comprising: an output unit that displays the cross-sectional image generated by the cross-sectional image generation unit on a display unit; and an acquisition unit that acquires a user's evaluation of the cross-sectional image output by the output unit.

11. The material design apparatus according to claim 10, further comprising a determination unit that determines whether the characteristics predicted to be possessed by the material of the cross-section shown in the cross-sectional image satisfy predetermined conditions, wherein the output unit causes the cross-sectional image whose characteristics are determined to satisfy predetermined conditions by the determination unit to be displayed on the display unit.

12. The material design apparatus according to claim 1 or 2, further comprising a display unit that displays at least one of the generation conditions, the cross-sectional image, the three-dimensional structure, the properties predicted to be possessed by the material of the cross-section shown in the cross-sectional image, and the properties predicted to be possessed by the material of the three-dimensional structure.

13. The material design apparatus according to claim 12, wherein the display unit displays, in a manner that allows for comparison, the predicted properties of the material in the cross-section shown in the cross-sectional image and the predicted properties of the material in the three-dimensional structure.

14. The material design apparatus according to claim 12, further comprising an operation receiving unit that receives an operation from a user to weight-combine at least two of the cross-sectional images displayed on the display unit, and the search unit sets the generation conditions based on the operation.

15. A material design method performed by a computer, comprising: a search step of setting generation conditions for generating a three-dimensional structure of a material, including information indicating at least one of the properties and structure of the material; a cross-sectional image generation step of generating a cross-sectional image of the material based on the generation conditions set by the search step; and a three-dimensional structure generation step of generating a three-dimensional structure of the material based on the cross-sectional image generated by the cross-sectional image generation step.

16. A program for causing a computer to execute the material design method described in claim 15.

Images