UV-Net Representations for Consistent 3D CAD Neural Processing
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional neural networks struggle to process unstructured B-rep data from 3D CAD objects due to their representation variability, leading to inconsistent results across different B-reps, hindering tasks like classification and recognition of 3D CAD objects.
Innovation Solution
Generate UV-net representations by discretizing parametric surfaces into 2D and 1D grids, extracting topological and geometric features as graphs, enabling efficient processing with neural networks.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If B-rep data is used to represent 3D CAD objects, then comprehensive geometric and topological information is captured, but the data becomes unstructured and difficult for neural networks to process
Solution Approach 1:
The patent segments B-rep data into distinct components: topological information (vertices, edges, faces) and geometric information (surface properties). This segmentation allows neural networks to process each component separately using appropriate architectures, resolving the contradiction between preserving comprehensive information and enabling easy processing.
Solution Approach 2:
The patent introduces an intermediary representation layer that transforms complex B-rep data into a format suitable for neural networks. This intermediary structure acts as a bridge, converting unstructured B-rep data into organized tensors while preserving geometric and topological relationships, thus enabling neural network processing without information loss.
2Adaptability or versatility
If different geometric entities are used to represent the same shape in B-reps, then flexibility in representation is achieved, but consistency of results across different B-reps becomes difficult
Solution Approach 1:
The patent applies parameter changes by normalizing geometric entities to a standard set of features. Different geometric representations (planes, NURBS surfaces, etc.) are transformed into a common parameter space with standardized attributes, ensuring that the same shape yields consistent results regardless of the original B-rep representation.
Solution Approach 2:
The patent creates a universal feature extraction framework that handles multiple types of geometric entities through a single unified process. This universal approach extracts consistent topological and geometric features from diverse B-rep representations, ensuring reliability across different input formats while maintaining adaptability.
3Extent of automation
If conventional neural networks are trained on B-rep data, then tasks like classification and recognition can be performed, but the training becomes impractical due to unstructured data
Solution Approach 1:
The patent applies preliminary action by preprocessing B-rep data into structured formats before neural network training. Topological entities and geometric features are extracted and organized into tensors in advance, making the data suitable for conventional neural networks. This preliminary structuring makes training practical while enabling automated classification and recognition tasks.
Data Source
AI summary
In various embodiments, a parameter domain graph application generates UV-net representations of 3D CAD objects for machine learning models. In operation, the parameter domain graph application generates a graph based on a B-rep of a 3D CAD object. The parameter domain graph application discretizes a parameter domain of a parametric surface associated with the B-rep into a 2D grid. The parameter domain graph application computes at least one feature at a grid point included in the 2D grid based on the parametric surface to generate a 2D UV-grid. Based on the graph and the 2D UV-grid, the parameter domain graph application generates a UV-net representation of the 3D CAD object. Advantageously, generating UV-net representations of 3D CAD objects that are represented using B-reps enables the 3D CAD objects to be processed efficiently using neural networks.