Multi-Format Shape Similarity Search Using ML Embeddings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
In computer-aided design and manufacturing, complex shapes represented in different formats pose challenges for characterization, comparison, and retrieval due to the inability to precisely describe geometries and attributes across various formats, leading to inefficiencies and potential loss of data representations.
Innovation Solution
A technique using trained machine learning models to generate embeddings for shapes across multiple formats, enabling the characterization of similarities and dissimilarities, and facilitating the retrieval of target shapes regardless of format, by determining embeddings for query shapes and matching them with shapes in different formats.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional text-based search techniques are used to retrieve CAD data, then the search process is simple to implement, but the precision and specificity of retrieving CAD data for complex shapes is insufficient
Solution Approach 1:
The patent replaces conventional text-based search mechanisms with a machine learning-based embedding system. Trained models convert CAD shapes into numerical embeddings that capture geometric characteristics, enabling precise similarity search through mathematical distance calculations rather than text matching. This substitution of mechanical search with intelligent embedding-based retrieval resolves the contradiction by achieving high precision without proportionally increasing system complexity.
Solution Approach 2:
The patent transforms CAD shape data from various formats into a unified embedding parameter space. By converting diverse format parameters (2D sketches, 3D models, meshes) into comparable numerical embeddings, the system enables precise similarity measurement. This parameter transformation allows the system to maintain search precision while managing complexity through standardized representation.
2Measurement precision
If manual mapping or format conversion is performed to compare objects in different formats, then attribute comparison accuracy is improved, but the time and computational resources required increase significantly
Solution Approach 1:
The patent extracts essential geometric attributes from CAD shapes in various formats and represents them as embeddings. Instead of performing complex manual mapping between formats, the system extracts key shape characteristics and encodes them into embeddings that preserve similarity relationships. This extraction approach maintains attribute comparison accuracy while dramatically reducing processing time by avoiding exhaustive format conversion.
Solution Approach 2:
The patent creates a universal embedding representation that works across multiple CAD formats simultaneously. The trained machine learning models can process 2D sketches, 3D models, meshes, and other formats through a unified embedding generation process. This multi-functional approach enables accurate attribute comparison without requiring separate processing pipelines for each format, thus reducing processing time while maintaining precision.
3Ease of operation
If complex geometries are described using words, phrases, or sentences for text-based search, then the search interface is user-friendly, but the precision and specificity of retrieving the correct CAD data is insufficient
Solution Approach 1:
The patent introduces embeddings as an intermediary between user queries and CAD data retrieval. The system processes user input (text descriptions, sketches, or images) into embeddings, which then serve as intermediaries to search the embedding database. This intermediary layer maintains ease of operation by accepting various user-friendly inputs while achieving high retrieval precision through embedding similarity matching, bridging the gap between simple interfaces and accurate results.
Data Source
Figure 1
Figure 2
Figure 3A
AI summary
One embodiment of the present invention sets forth a technique for analyzing similarities associated with a plurality of shapes. The technique includes determining a first embedding for a first query shape associated with a first format and a first plurality of embeddings for a first plurality of shapes associated with a second format, wherein the first embedding and the first plurality of embeddings are generated by one or more trained machine learning models based on the first query shape and the first plurality of shapes. The technique also includes matching, based on the first embedding and the first plurality of embeddings, the first query shape to one or more shapes included in the first plurality of shapes. The technique further includes outputting the one or more shapes in a first response associated with the first query shape.