Neural Network Parameterization for 3D Shape Optimization
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing shape optimization methods in computer-aided design and engineering are computationally demanding, limiting the search space and efficiency due to high computational requirements, often necessitating simplifications such as reducing degrees of freedom or restricting shapes to those with few parameters.
Innovation Solution
A machine-learning based method using a neural network that learns parameterization vectors for 3D modeled objects, allowing for differentiable simulation-based shape optimization by minimizing a loss function that penalizes disparities between network outputs and actual object representations, enabling efficient optimization in a lower-dimensional space.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If existing shape optimization methods are used, then optimization can be performed, but computational requirements are high and search space is limited
Solution Approach 1:
The patent transforms the high-dimensional shape optimization problem into a low-dimensional latent space problem. By learning a parameterization that maps from a small number of latent variables to complex 3D shapes, the method enables efficient optimization in this compressed dimensionality while maintaining the ability to represent complex geometries, thus reducing computational resource consumption and improving optimization productivity.
2Adaptability or versatility
If the search space is expanded to include more complex shapes, then design flexibility improves, but computational complexity increases
Solution Approach 1:
The patent introduces a learned parameterization model as an intermediary between the simple latent space and complex shape representations. This intermediary enables the system to work with complex shapes and expanded design flexibility while maintaining computational efficiency by performing optimizations in the simpler latent space, thus resolving the contradiction between adaptability and computational complexity.
3Productivity
If degrees of freedom are reduced to simplify optimization, then computational efficiency improves, but shape representation accuracy decreases
Solution Approach 1:
The patent fundamentally changes the parameterization approach by learning an optimal mapping from latent variables to shape parameters. Instead of using traditional reduced-degree-of-freedom parameterizations that compromise accuracy, the learned parameterization efficiently encodes complex shapes in a low-dimensional latent space while preserving shape representation accuracy, thus achieving both optimization efficiency and manufacturing precision.
Data Source
AI summary
A computer-implemented method of machine-learning. The method comprises providing a dataset of 3D modeled objects each representing a mechanical part. Each 3D modeled object comprises a specification of a geometry of the mechanical part. The method further comprises learning a set of parameterization vectors each respective to a respective 3D modeled object of the dataset and a neural network configured to take as input a parameterization vector and to output a representation of a 3D modeled object usable in a differentiable simulation-based shape optimization. The learning comprises minimizing a loss that penalizes, for each 3D modeled object of the dataset, a disparity between the output of the neural network for an input parameterization vector respective to the 3D modeled object and a representation of the 3D modeled object. The representation of the 3D modeled object is usable in a differentiable simulation-based shape optimization.