Neural Network Inference for Solid CAD Features from Freehand Drawings
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for inferring 3D modeled objects from freehand drawings are limited by the generation of meshes that are difficult to edit and lack semantics, requiring annotated strokes, specific user inputs, and assuming restricted plane orientations, which restricts user flexibility and accuracy.
Innovation Solution
A computer-implemented method using a neural network that learns to infer a solid CAD feature from a freehand drawing, comprising recurrent neural networks and convolutional neural networks, which processes the drawing to output data for inferring a sequence of curves and sweep operations, allowing for unsupervised and supervised training to generate a compact and editable 3D model.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If mesh generation methods are used to infer 3D objects from freehand drawings, then 3D shape reconstruction is achieved, but the resulting meshes are difficult to edit and lack semantic meaning
Solution Approach 1:
The patent uses neural networks to learn the mapping from 2D freehand drawings to 3D CAD features by training on paired datasets. Instead of directly generating meshes, the system learns to copy the semantic structure of CAD features from training examples, enabling accurate reconstruction while maintaining editability through parameterized feature representations
Solution Approach 2:
The patent replaces traditional geometric reconstruction algorithms with machine learning-based neural networks. The system substitutes deterministic geometric processing with probabilistic learning models that can infer CAD features directly from sketch data, achieving both accuracy and semantic meaning through pattern recognition rather than mechanical geometry processing
2Ease of manufacture
If traditional sketch-based methods are used, then 3D modeling is achieved, but annotated strokes and specific user inputs are required
Solution Approach 1:
The neural network system performs automatic inference of CAD features from freehand drawings without requiring user annotations or additional inputs. The trained model self-service completes the 3D reconstruction process by independently interpreting sketch geometry and inferring corresponding CAD features, eliminating the need for user-guided annotation or interaction
Solution Approach 2:
The patent performs preliminary training of neural networks on large datasets of paired sketch-CAD feature examples before deployment. This preliminary learning action enables the system to acquire the knowledge needed for automatic feature inference, so that during actual use, no additional user inputs or annotations are required as the model has already learned the mapping relationships
3Device complexity
If methods assuming restricted plane orientations are used, then processing simplicity is maintained, but user flexibility is reduced
Solution Approach 1:
The patent employs recurrent neural networks (RNNs) that process sketch data in a sequential, dynamic manner rather than assuming fixed geometric constraints. The RNN architecture allows the system to adapt to various sketch orientations and complexities by processing information step-by-step, maintaining algorithmic simplicity while achieving versatility through the dynamic nature of sequential processing
Data Source
AI summary
The disclosure notably relates to computer-implemented method for learning a neural network configured for inference, from a freehand drawing representing a 3D shape, of a solid CAD feature representing the 3D shape. The method includes providing a dataset including freehand drawings each representing a respective 3D shape, and learning the neural network based on the dataset. The method forms an improved solution for inference, from a freehand drawing representing a 3D shape, of a 3D modeled object representing the 3D shape.