Generative Autoencoder for Hierarchical CAD Functional Structures
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for learning 3D shapes in CAD systems fail to effectively handle hierarchical structured data, leading to unrealistic results, as they do not account for the fundamental characteristic of 3D shapes in CAD data, such as mechanical assemblies of rigid parts.
Innovation Solution
A generative autoencoder is developed to learn and generate functional structures, represented as trees with leaf nodes for rigid parts and non-leaf nodes for mechanical links and duplications, using an encoder and decoder with recursive encoding and decoding processes to capture hierarchical relationships and generate realistic mechanical assemblies.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If existing methods represent 3D shapes with mesh, point clouds, voxels or multi-view representations, then the shapes can be processed, but the hierarchical structure of CAD data is not captured and unrealistic results are produced
Solution Approach 1:
The patent segments the CAD data into a hierarchical tree structure where the root node represents the entire assembly and child nodes represent sub-assemblies or individual parts. This segmentation preserves the hierarchical relationships and structural information that are lost in traditional voxel or mesh representations, enabling realistic generation of CAD data by maintaining the parent-child relationships and assembly constraints.
2Adaptability or versatility
If deep neural networks are applied to 3D shape analysis using voxel grids, then computer vision techniques can be extended, but hierarchically structured CAD data cannot be handled
Solution Approach 1:
The patent transitions from representing 3D shapes in spatial dimensions (voxels, meshes) to representing them in a hierarchical dimension using tree structures. This dimensional transformation allows deep neural networks to process CAD data while preserving hierarchical relationships, combining the adaptability of deep learning with the reliability needed for hierarchical data through a different representational framework.
3Ease of manufacture
If CAD systems use traditional 3D shape representations, then processing can be performed, but the fundamental characteristic of hierarchical structure is not accounted for
Solution Approach 1:
The patent implements a nested hierarchical tree structure where parent nodes contain child nodes, similar to nested dolls. The root node nests sub-assembly nodes, which in turn nest individual part nodes. This nesting structure preserves the hierarchical information at each level while maintaining processing capability, allowing the system to work with CAD data at different levels of detail without losing the fundamental hierarchical characteristics.
Data Source
AI summary
The disclosure notably relates to a computer-implemented method for teaching a generative autoencoder. The generative autoencoder is configured to generate functional structures. A functional structure is a data structure representing a mechanical assembly of rigid parts and which includes a tree. Each leaf node represents a shape and positioning of a respective rigid part and a force exerted on the respective rigid part. Each non-leaf node with several children represents a mechanical link between sub-assemblies. Each sub-assembly is represented by a respective one of the several children. Each non-leaf node with a single child represents a duplication of the sub-assembly represented by the single child. The method includes obtaining a dataset including functional structures. The method further comprises teaching the generative autoencoder on the dataset. This constitutes an improved method for teaching a generative autoencoder configured for generating functional structures.