Neural Network Dataset for Editable Feature Tree Inference
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for inferring a plausible editable feature tree from a discrete geometrical representation of a 3D shape are inefficient, particularly when dealing with complex geometries, as they often fail to reconstruct valid boundary representations or propagate errors, and lack accuracy in supervised learning approaches.
Innovation Solution
A computer-implemented method forms a dataset for learning a neural network that infers an editable feature tree from a discrete geometrical representation of a 3D shape, using a tree arrangement of geometrical operations applied to leaf geometrical shapes, with data pieces including both the editable feature tree and its corresponding discrete geometrical representation, and employs a beam search algorithm for refining the inferred tree to improve accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If current methods are used to infer editable feature trees from discrete geometrical representations, then the process can be automated, but accuracy deteriorates particularly with complex geometries
Solution Approach 1:
The method segments the inference process into distinct stages: discrete geometrical representation processing, neural network inference, and beam search refinement. This segmentation allows each stage to be optimized independently, with the beam search algorithm specifically addressing accuracy improvements in the final refinement stage without compromising the automation of the overall process.
2Productivity
If traditional reconstruction methods are applied to complex geometries, then the process can be completed, but reliability deteriorates due to failure to reconstruct valid boundary representations
Solution Approach 1:
The beam search refinement algorithm implements a feedback mechanism where the inferred feature tree is evaluated against the original discrete geometrical representation, and adjustments are made iteratively to improve reliability. This feedback loop ensures that the final output maintains validity while preserving the productivity of the automated reconstruction process.
3Loss of time
If supervised learning approaches are used with limited training data, then the model can be trained quickly, but accuracy deteriorates
Solution Approach 1:
The method performs preliminary actions by pre-processing the discrete geometrical representations into standardized formats and pre-defining the feature tree structure before neural network inference. This preliminary preparation reduces the complexity of the learning task, allowing for faster training with limited data while maintaining or improving accuracy through the structured approach.
Data Source
AI summary
The disclosure notably relates to a computer-implemented method for forming a dataset configured for learning a neural network. The neural network is configured for inference, from a discrete geometrical representation of a 3D shape, of an editable feature tree representing the 3D shape. The editable feature tree comprises a tree arrangement of geometrical operations applied to leaf geometrical shapes. The method includes obtaining respective data pieces, and inserting a part of the data pieces in the dataset each as a respective training sample. The respective 3D shape of each of one or more first data pieces inserted in the dataset is identical to the respective 3D shape of respective one or more second data pieces not inserted in the dataset. The method forms an improved solution for digitization.