3D Shape Embeddings for Generative Design Exploration
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing computer-aided design (CAD) tools face challenges in efficiently comparing and exploring large numbers of complex 3D designs due to limitations in computational representations for shapes, making it difficult for users to identify aesthetically preferred designs within generative design spaces.
Innovation Solution
A computer-implemented method using convolutional neural networks (CNNs) to generate shape embeddings that accurately represent 3D geometries, allowing for efficient comparisons and exploration of design spaces based on aesthetic preferences by training an autoencoder to map view sets into fixed-size shape embeddings.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If traditional computational representations for shapes are used in CAD tools, then existing CAD functionality is maintained, but efficient comparison of large numbers of complex 3D designs is precluded due to large size and high computational requirements
Solution Approach 1:
The patent transforms the computational representation of 3D shapes by changing parameters from traditional large-size representations (e.g., detailed mesh models) to compact shape embeddings that capture essential geometric features. This parameter transformation enables efficient comparison of thousands of designs while maintaining sufficient accuracy for aesthetic evaluation.
Solution Approach 2:
The patent extracts the essential shape characteristics from complex 3D geometries by using neural network encoders to generate condensed shape embeddings. This extraction process separates the critical geometric features needed for comparison from the unnecessary detailed representation data, enabling efficient design space exploration.
2Measurement precision
If existing shape representation methods are used, then computational representations are available, but they do not capture complex shapes sufficiently for accurate comparison
Solution Approach 1:
The patent changes the parameter representation from detailed geometric data to compressed shape embeddings that preserve essential shape characteristics. The neural network encoding process transforms high-dimensional shape data into lower-dimensional embeddings that maintain sufficient fidelity for accurate aesthetic comparison while reducing data volume.
Solution Approach 2:
The patent creates simplified copies of complex 3D shapes in the form of shape embeddings. These embedding representations serve as efficient proxies that capture the essential shape information needed for comparison without requiring the full complexity of the original geometries.
3Ease of operation
If manual screening of designs is performed by users, then aesthetic preferences can be evaluated, but the process is time-consuming and tedious for hundreds or thousands of designs
Solution Approach 1:
The patent replaces the manual mechanical process of visual inspection with an automated computational system using neural networks. The shape embedding generation and comparison algorithms automatically evaluate aesthetic preferences for thousands of designs, substituting human visual analysis with machine-based geometric feature comparison.
Solution Approach 2:
The patent enables the design exploration system to serve itself by automatically generating shape embeddings and performing comparisons without requiring manual user intervention for each design evaluation. The system autonomously processes large numbers of designs through the embedded neural network pipeline.
Data Source
AI summary
In various embodiments, a training application generates a trained encoder that automatically generates shape embeddings having a first size and representing three-dimensional (3D) geometry shapes. First, the training application generates a different view activation for each of multiple views associated with a first 3D geometry based on a first convolutional neural network (CNN) block. The training application then aggregates the view activations to generate a tiled activation. Subsequently, the training application generates a first shape embedding having the first size based on the tiled activation and a second CNN block. The training application then generates multiple re-constructed views based on the first shape embedding. The training application performs training operation(s) on at least one of the first CNN block and the second CNN block based on the views and the re-constructed views to generate the trained encoder.