Microstructures using generative adversarial networks

Patent metadata

Field	Value
Publication number	US-11501037-B2
Application number	US-201916441483-A
Country	US
Kind code	B2
Filing date	Jun 14, 2019
Priority date	Jun 14, 2019
Publication date	Nov 15, 2022
Grant date	Nov 15, 2022

Abstract

A method for designing microstructures includes receiving at least one material property constraint for a design of at least one microstructure, the at least one microstructure configured to be a part of a larger macrostructure. At least one neighborhood connectivity constraint for the design of the at least one microstructure is received. One or more designs of the at least one microstructure is generated using a generative adversarial network (GAN) that is based on the at least one material property constraint and the at least one neighborhood connectivity constraint.

First claim

What is claimed is:

1. A method designing microstructures, comprising:
receiving at least one material property constraint for a design of at least one microstructure, the at least one microstructure configured to be a part of a larger macrostructure;
receiving at least one neighborhood connectivity constraint for the design of the at least one microstructure; and
generating one or more designs of the at least one microstructure using a generative adversarial network (GAN) based on the at least one material property constraint and the at least one neighborhood connectivity constraint, wherein the GAN comprises a multi-conditional GAN that controls both the at least one material property and the at least one neighborhood connectivity constraint using context-based synthesis derived from computer vision applications.

2. The method of claim 1 , wherein the at least one neighborhood connectivity constraints comprise substantially identical interfaces at neighborhood boundaries.

3. The method of claim 1 , further comprising training the GAN with one or more training microstructures.

4. The method of claim 1 , wherein the GAN is configured to produce the one or more designs by inverse design of microstructures through one or more of physical test data and simulated data.

5. The method of claim 1 wherein the GAN is configured to produce the one or more designs on the fly as part of a macro-scale design optimization algorithm where the material property constraints are determined.

6. The method of claim 1 , wherein the one or more microstructures comprise small-scale assemblies with one or more different macro-scale properties than their base materials.

7. A system for designing microstructures, comprising:
a processor; and
a memory coupled to the processor, the memory storing computer executable instructions, that, when executed by the processor, cause the system to:
receive at least one material property constraint for a design of at least one microstructure, the at least one microstructure configured to be a part of a larger macrostructure;
receive at least one neighborhood connectivity constraint for the design of the at least one microstructure; and
generate one or more designs of the at least one microstructure using a generative adversarial network (GAN) based on the at least one material property constraint and the at least one neighborhood connectivity constraint, wherein the GAN comprises a multi-conditional GAN that controls both the at least one material property and the at least one neighborhood connectivity constraint using context-based synthesis derived from computer vision applications.

8. The system of claim 7 , wherein the at least one neighborhood connectivity constraints comprise substantially identical interfaces at neighborhood boundaries.

9. The system of claim 7 , wherein the processor is configured to train the GAN with one or more training microstructures.

10. The system of claim 7 , wherein the GAN is configured to produce the one or more designs by inverse design of microstructures through one or more of physical test data and simulated data.

11. The system of claim 7 wherein the GAN is configured to produce the one or more designs on the fly as part of a macro-scale design optimization algorithm where the material property constraints are determined.

12. The system of claim 7 , wherein the one or more microstructures comprise small-scale assemblies with one or more different macro-scale properties than their base materials.

13. A non-transitory computer readable medium storing computer program instructions for designing microstructures, the computer program instructions when executed by a processor cause the processor to perform operations comprising:
receiving at least one material property constraint for a design of at least one microstructure, the at least one microstructure configured to be a part of a larger macrostructure;
receiving at least one neighborhood connectivity constraint for the design of the at least one microstructure; and
generating one or more designs of the at least one microstructure using a generative adversarial network (GAN) based on the at least one material property constraint and the at least one neighborhood connectivity constraint, wherein the GAN comprises a multi-conditional GAN that controls both the at least one material property and the at least one neighborhood connectivity constraint using context-based synthesis derived from computer vision applications.

14. The non-transitory computer readable medium of claim 13 , wherein the at least one neighborhood connectivity constraints comprise substantially identical interfaces at neighborhood boundaries.

15. The non-transitory computer readable medium of claim 13 , further comprising training the GAN with one or more training microstructures.

16. The non-transitory computer readable medium of claim 13 , wherein the GAN is configured to produce the one or more designs by inverse design of microstructures through one or more of physical test data and simulated data.

17. The non-transitory computer readable medium of claim 13 , wherein the GAN is configured to produce the one or more designs on the fly as part of a macro-scale design optimization algorithm where the material property constraints are determined.

18. The non-transitory computer readable medium of claim 13 , wherein generating the one or more designs of the at least one microstructure using the GAN comprises solving an inverse design problem of determining target material properties for the at least one microstructure, the target material properties formulated as a learning problem via the GAN.

19. The method of claim 1 , wherein generating the one or more designs of the at least one microstructure using the GAN comprises solving an inverse design problem of determining target material properties for the at least one microstructure, the target material properties formulated as a learning problem via the GAN.

20. The system of claim 7 , wherein generating the one or more designs of the at least one microstructure using the GAN comprises solving an inverse design problem of determining target material properties for the at least one microstructure, the target material properties formulated as a learning problem via the GAN.

Assignees

• Palo Alto Res Ct Inc

Inventors

• Gecer Ulu Nurcan
• Korneev Svytoslav
• Nelaturi Saigopal

Classifications

• G06N3/047

  Probabilistic or stochastic networks · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/088

  Non-supervised learning, e.g. competitive learning · CPC title

• G06F30/27 Primary

  using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model · CPC title

• G05B2219/40534

  Generate derivative, change of position error · CPC title