Granted Patent B2

US 12,265,765 · App. 18/908,028 · Granted Apr 1, 2025

Inventors: Maor Farid (Rosh HaAyin, IL); Mordechai Moravia (Rosh HaAyin, IL)

Assignee: SCINTIUM LTD

G06F30/27G06F40/20

Abstract

Claims (46)

1. A computerized method for training an artificial intelligence (AI) model to generate a 3D Computer-Aided Design (CAD), the method comprising:

obtaining input data comprising at least requirements for a new design; and

training the AI model to interpret the input data and generate a corresponding CAD design, wherein the CAD design is an assembly of at least two separate components,

wherein training the AI model further comprises training the AI model to generate a CAD design comprising at least one known component;

wherein training the AI model further comprises training the AI model to:

generate an embedding vector corresponding to an ideal component to be included in the CAD design;

access a vector database storing a plurality of vectors, wherein each vector represents a known component used in designs;

search the vector database to identify a matching vector to the embedding vector;

retrieve a known component corresponding to the matching vectors; and

generate the CAD design using the retrieved known component; and

wherein training the AI model further comprises training the AI model to identify and model engineering constraints between the different components by masking some of the components during training and teaching the AI model to predict an appropriate component to place in the masked position.

2. The method of claim 1 , wherein each of the components is selected from a sub-assembly and a part.

3. The method of claim 1 , wherein the input comprises a plurality of input modalities, wherein the method further comprises:

prior to training the AI model, preprocessing the plurality of input modalities, using at least a Large Language Model (LLM), by:

generating embedding vectors for each of the modalities using vector embedding techniques; and

fusing the embedding vectors in various combinations of the modalities to create linked representations of the input data modalities; and

training the AI model to interpret the input data and the generated combination.

4. The method of claim 3 , wherein each of the input modalities is selected from a group comprising: a CAD design and requirements pertaining to the CAD design, a textual description, a sketch, a 2D illustration image, and a technical specification.

5. The method of claim 1 , further comprising enriching the input data with targeted organization-specific data.

6. The method of claim 1 , wherein enriching the input comprises integrating data from the organization's Product Lifecycle Management (PLM) system, including CAD data stored within the PLM system and the organization's documents.

7. The method of claim 1 , further comprising training the AI model to:

enrich the input based on at least responses to requests for additional information based on pre-trained engineering requirements or data based on responses to user-initiated questions; and

process the enriched input to generate the CAD design.

8. The method of claim 1 , wherein the AI model is trained to generate a corresponding CAD design compliant with Design for Manufacturing and Automation (DFMA) standards.

9. The method of claim 1 , wherein the AI model is trained to output technical specification of the outputted CAD design.

10. A computer system for training an artificial intelligence (AI) model to generate a 3D Computer-Aided Design (CAD), the system comprising a processing circuitry being configured to:

obtain input data comprising at least requirements for a new design; and

train the AI model to interpret the input data and generate a corresponding CAD design, wherein the CAD design is an assembly of at least two separate components,

wherein training the AI model further comprises training the AI model to generate a CAD design comprising at least one known component;

wherein training the AI model further comprises training the AI model to:

generate an embedding vector corresponding to an ideal component to be included in the CAD design;

access a vector database storing a plurality of vectors, wherein each vector represents a known component used in designs;

search the vector database to identify a matching vector to the embedding vector;

retrieve a known component corresponding to the matching vectors; and

generate the CAD design using the retrieved known component; and

wherein the processing circuitry being configured to train the AI model to identify and model engineering constraints between the different components by masking some of the components during training and teaching the AI model to predict an appropriate component to place in the masked position.

11. A non-transitory computer readable storage medium tangibly embodying a program of instructions that, when executed by a computer, cause the computer to perform a computerized method for training an artificial intelligence (AI) model to generate a 3D Computer-Aided Design (CAD), the method comprising:

obtaining input data comprising at least requirements for a new design; and training the AI model to interpret the input data and generate a corresponding CAD design, wherein the CAD design is an assembly of at least two separate components,

wherein training the AI model further comprises training the AI model to generate a CAD design comprising at least one known component;

wherein training the AI model further comprises training the AI model to:

generate an embedding vector corresponding to an ideal component to be included in the CAD design;

access a vector database storing a plurality of vectors, wherein each vector represents a known component used in designs;

search the vector database to identify a matching vector to the embedding vector;

retrieve a known component corresponding to the matching vectors; and

generate the CAD design using the retrieved known component; and

wherein training the AI model further comprises training the AI model to identify and model engineering constraints between the different components by masking some of the components during training and teaching the AI model to predict an appropriate component to place in the masked position.

Assignments (2)

Continuity (4)

References Cited (34)