# US 2026/0017422 A1 · Methods for refining topology optimized designs of structures and non-transitory computer-readable media associated therewith

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2026/0017422 A1
- Document ID: US20260017422A1
- Status: Published application
- Kind: A1
- Application number: 18/768,623
- Filed: 2024-07-10
- Published: 2026-01-15
- Earliest claimed: 2024-07-10
- Assignees: The Boeing Company
- Applicants: The Boeing Company
- Inventors: Brian S. Smith
- Web page: https://patentgrep.com/patent/US20260017422A1

## CPC

- G06F 30/15: Vehicle, aircraft or watercraft design
- G06F 30/27: using machine learning, e.g. artificial intelligence, neural networks, support vector machines [SVM] or training a model
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)

## Abstract

A method for refining a topology optimized design of a structure includes receiving a source image file representative of the topology optimized design for the structure at a latent space diffusion model with a network control extension, generating a natural language conditioning prompt to describe desired content for a refined design of the structure, preprocessing the source image file at the network control extension using feature extraction tools to extract image features from the source image file, generating a conditioning image file at the network control extension, applying the natural language conditioning prompt, the conditioning image file and a series of noisy image files to a neural network of the latent space diffusion model, and processing the natural language conditioning prompt, the conditioning image file and the series of noisy image files through the neural network using a reverse diffusion process to create an intermediate generative image file.

## Claims

### Claim 1

A method for refining a topology optimized design of a structure, comprising:
receiving a source image file representative of the topology optimized design for the structure at a latent space diffusion model with a network control extension;
generating a natural language conditioning prompt to describe desired content for a refined design of the structure based on the topology optimized design;
preprocessing the source image file at the network control extension using feature extraction tools to extract image features from the source image file;
generating a conditioning image file at the network control extension based on the image features extracted from the source image file;
applying the natural language conditioning prompt, the conditioning image file and a series of noisy image files to a neural network of the latent space diffusion model; and
processing the natural language conditioning prompt, the conditioning image file and the series of noisy image files through the neural network using a reverse diffusion process to create an intermediate generative image file.

### Claim at position 2

2-7. (canceled)

### Claim 8 (depends on 1)

The method of claim 1 wherein the image features comprise at least one of pose features, background features, foreground features, depth features, edge features, line features, straight-line features, object features, texture features, color features and transparency features.

### Claim at position 4

9-11. (canceled)

### Claim 12 (depends on 1)

The method of claim 1, further comprising:
preprocessing the source image file at the latent space diffusion model using a forward diffusion process to obtain the series of noisy image files in which a level of noise in the series of noisy image files ranges from less noise in a first noisy image file to more noise in successive image files such that a last noisy image file in the series includes a highest amount of noise.

### Claim 13

(canceled)

### Claim 14 (depends on 1)

The method of claim 1, further comprising:
iteratively processing the intermediate generative image file, the natural language conditioning prompt and the conditioning image file through the neural network to refine the intermediate generative image file until the intermediate generative image file is representative of the desired content for the refined design of the structure.

### Claim 15 (depends on 14)

The method of claim 14, further comprising:
generating a refined generative image file representative of the refined design for the structure at the latent space diffusion model based on a final iteration of the intermediate generative image file.

### Claim at position 9

16-18. (canceled)

### Claim 19 (depends on 15)

The method of claim 15 wherein the refined generative image file comprises an exploded view of the structure that shows at least two parts used to fabricate the structure in a disassembled representation.

### Claim at position 11

20-21. (canceled)

### Claim 22 (depends on 1)

The method of claim 1, further comprising:
processing the intermediate generative image file, the natural language conditioning prompt and the conditioning image file through the neural network in a non-iterative manner to refine the intermediate generative image file until the intermediate generative image file is representative of the desired content for the refined design of the structure.

### Claim 23 (depends on 1)

The method of claim 1, further comprising:
comparing the intermediate generative image file to the source image file and the desired content;
generating a second natural language conditioning prompt to describe further desired content for the refined design of the structure based on the comparing; and
iteratively processing the intermediate generative image file, the second natural language conditioning prompt and the conditioning image file through the neural network to refine the intermediate generative image file until the intermediate generative image file is representative of the further desired content for the refined design of the structure.

### Claim 24 (depends on 1)

The method of claim 1, further comprising:
generating natural language conditioning prompts to describe further desired content for the refined design of the structure based on intermediate results; and
refining design inputs based on iterative processing of the intermediate generative image file, the natural language conditioning prompt and the conditioning image file through the neural network until the intermediate generative image file is representative of the further desired content for the refined design of the structure.

### Claim 25 (depends on 1)

The method of claim 1, further comprising:
comparing the intermediate generative image file to the source image file and the desired content;
preprocessing the source image file at the network control extension using the feature extraction tools to extract second image features from the source image file;
generating a second conditioning image file at the network control extension based on the second image features extracted from the source image file; and
iteratively processing the intermediate generative image file, the natural language conditioning prompt and the second conditioning image file through the neural network to refine the intermediate generative image file until the intermediate generative image file is representative of the desired content for the refined design of the structure.

### Claim 26 (depends on 1)

The method of claim 1, further comprising:
generating conditioning images based on intermediate results; and
refining design inputs based on iterative processing of the intermediate generative image file, the natural language conditioning prompt and the conditioning image file through the neural network until the intermediate generative image file is representative of the desired content for the refined design of the structure.

### Claim 27 (depends on 1)

The method of claim 1, further comprising:
comparing the intermediate generative image file to the source image file and the desired content;
preprocessing the intermediate generative image file at the network control extension using the feature extraction tools to extract second image features from the intermediate generative image file;
generating a second conditioning image file at the network control extension based on the second image features extracted from the intermediate generative image file; and
iteratively processing the intermediate generative image file, the natural language conditioning prompt and the second conditioning image file through the neural network to refine the intermediate generative image file until the intermediate generative image file is representative of the desired content for the refined design of the structure.

### Claim 28

A method for refining a topology optimized design of a structure, comprising:
receiving a source image file representative of the topology optimized design for the structure at a latent space diffusion model with a network control extension;
generating a natural language conditioning prompt to describe desired content for a refined design of the structure based on the topology optimized design;
preprocessing the source image file at the network control extension using feature extraction tools to extract at least one image feature from the source image file;
generating a conditioning image file at the network control extension based on the at least one image feature extracted from the source image file;
applying the natural language conditioning prompt, the conditioning image file and a series of noisy image files to a neural network of the latent space diffusion model, wherein the series of noisy image files is related to the source image file; and
processing the natural language conditioning prompt, the conditioning image file and the series of noisy image files through the neural network using a reverse diffusion process to create an intermediate generative image file.

### Claim at position 19

29-36. (canceled)

### Claim 37 (depends on 28)

The method of claim 28, further comprising:
iteratively processing the intermediate generative image file, the natural language conditioning prompt and the conditioning image file through the neural network to refine the intermediate generative image file until the intermediate generative image file is representative of the desired content for the refined design of the structure; and
generating a refined generative image file representative of the refined design for the structure at the latent space diffusion model based on a final iteration of the intermediate generative image file.

### Claim 38 (depends on 37)

The method of claim 37, further comprising:
receiving the refined generative image file representative of the refined design for the structure at the latent space diffusion model with the network control extension;
generating a second natural language conditioning prompt to describe desired content for an assembly within the structure based on the refined generative image file;
preprocessing the refined generative image file at the network control extension using the feature extraction tools to extract at least one image feature from the refined generative image file;
generating a second conditioning image file at the network control extension based on the at least one image feature extracted from the refined generative image file;
applying the second natural language conditioning prompt, the second conditioning image file and a second series of noisy image files to the neural network of the latent space diffusion model, wherein the second series of noisy image files is related to the refined generative image file; and
processing the second natural language conditioning prompt, the second conditioning image file and the second series of noisy image files through the neural network using the reverse diffusion process to create an intermediate assembly image file.

### Claim 39 (depends on 38)

The method of claim 38, further comprising:
iteratively processing the intermediate assembly image file, the second natural language conditioning prompt and the second conditioning image file through the neural network to refine the intermediate assembly image file until the intermediate assembly image file is representative of the desired content for the assembly within the structure; and
generating a generative assembly image file representative of the assembly at the latent space diffusion model based on the intermediate assembly image file.

### Claim at position 23

40-43. (canceled)

### Claim 44 (depends on 28)

The method of claim 28, further comprising:
receiving the intermediate generative image file representative of the refined design for the structure at the latent space diffusion model with the network control extension;
generating a second natural language conditioning prompt to describe desired content for an assembly within the structure based on the intermediate generative image file;
preprocessing the intermediate generative image file at the network control extension using the feature extraction tools to extract at least one image feature from the intermediate generative image file;
generating a second conditioning image file at the network control extension based on the at least one image feature extracted from the intermediate generative image file;
applying the second natural language conditioning prompt, the second conditioning image file and a second series of noisy image files to the neural network of the latent space diffusion model, wherein the second series of noisy image files is related to the intermediate generative image file; and
processing the second natural language conditioning prompt, the second conditioning image file and the second series of noisy image files through the neural network using the reverse diffusion process to create an intermediate assembly image file.

### Claim 45

(canceled)

### Claim 46 (depends on 28)

The method of claim 28, further comprising:
receiving the intermediate generative image file representative of the refined design for the structure at the latent space diffusion model with the network control extension;
generating a third natural language conditioning prompt to describe desired content for a subassembly within an assembly of the structure based on the intermediate generative image file;
preprocessing the intermediate generative image file at the network control extension using the feature extraction tools to extract at least one image feature from the intermediate generative image file;
generating a third conditioning image file at the network control extension based on the at least one image feature extracted from the intermediate generative image file;
applying the third natural language conditioning prompt, the third conditioning image file and a third series of noisy image files to the neural network of the latent space diffusion model, wherein the third series of noisy image files is related to the intermediate generative image file; and
processing the third natural language conditioning prompt, the third conditioning image file and the third series of noisy image files through the neural network using the reverse diffusion process to create an intermediate subassembly image file.

### Claim 47

(canceled)

### Claim 48 (depends on 28)

The method of claim 28, further comprising:
receiving the intermediate generative image file representative of the refined design for the structure at the latent space diffusion model with the network control extension;
generating a fourth natural language conditioning prompt to describe desired content for a part within a subassembly of an assembly of the structure based on the intermediate generative image file;
preprocessing the intermediate generative image file at the network control extension using the feature extraction tools to extract at least one image feature from the intermediate generative image file;
generating a fourth conditioning image file at the network control extension based on the at least one image feature extracted from the intermediate generative image file;
applying the fourth natural language conditioning prompt, the fourth conditioning image file and a fourth series of noisy image files to the neural network of the latent space diffusion model, wherein the fourth series of noisy image files is related to the intermediate generative image file; and
processing the fourth natural language conditioning prompt, the fourth conditioning image file and the fourth series of noisy image files through the neural network using the reverse diffusion process to create an intermediate part image file.

### Claim 49

(canceled)

### Claim 50

A non-transitory computer-readable medium comprising program instructions that, when executed by at least one processor, cause at least one computing device to perform a method for refining a topology optimized design of a structure, the method comprising:
receiving a source image file representative of the topology optimized design for the structure at a latent space diffusion model with a network control extension;
generating a natural language conditioning prompt to describe desired content for a refined design of the structure based on the topology optimized design;
preprocessing the source image file at the network control extension using feature extraction tools to extract image features from the source image file;
generating a conditioning image file at the network control extension based on the image features extracted from the source image file;
applying the natural language conditioning prompt, the conditioning image file and a series of noisy image files to a neural network of the latent space diffusion model; and
processing the natural language conditioning prompt, the conditioning image file and the series of noisy image files through the neural network using a reverse diffusion process to create an intermediate generative image file.

### Claim at position 31

51-56. (canceled)
