BRepNet: A topological message passing system for solid models (2104.00706v2)

Abstract: Boundary representation (B-rep) models are the standard way 3D shapes are described in Computer-Aided Design (CAD) applications. They combine lightweight parametric curves and surfaces with topological information which connects the geometric entities to describe manifolds. In this paper we introduce BRepNet, a neural network architecture designed to operate directly on B-rep data structures, avoiding the need to approximate the model as meshes or point clouds. BRepNet defines convolutional kernels with respect to oriented coedges in the data structure. In the neighborhood of each coedge, a small collection of faces, edges and coedges can be identified and patterns in the feature vectors from these entities detected by specific learnable parameters. In addition, to encourage further deep learning research with B-reps, we publish the Fusion 360 Gallery segmentation dataset. A collection of over 35,000 B-rep models annotated with information about the modeling operations which created each face. We demonstrate that BRepNet can segment these models with higher accuracy than methods working on meshes, and point clouds.

• The paper presents a novel BRepNet architecture that applies convolution on oriented coedges to leverage rich topological data in CAD models.
• It introduces the comprehensive Fusion 360 Gallery dataset with over 35,000 annotated B-rep models to support CAD-based machine learning research.
• Empirical results demonstrate that BRepNet outperforms traditional mesh and point cloud methods, achieving a segmentation accuracy of 92.52% and an IoU of 77.10%.

BRepNet: A Topological Message Passing System for Solid Models

Overview

BRepNet is introduced as a neural network architecture designed to work directly with boundary representation (B-rep) data structures, standard in CAD applications. This approach operates without resorting to approximations as meshes or point clouds, leveraging the intricate topological details found within B-reps. The network implements convolutional operations on oriented coedges, exploiting local topological information to enhance feature detection. Notably, BRepNet demonstrates superior segmentation accuracy over traditional mesh and point cloud methods.

Contributions and Results

The paper's principal contributions are threefold:

1. BRepNet Architecture: A novel convolution technique capitalizes on the B-rep's topological information. By structuring convolutional kernels concerning coedges, the architecture efficiently processes B-rep data, a unique ability that distinguishes it from graph-based techniques.
2. Fusion 360 Gallery Dataset: Introduction of a comprehensive dataset comprising over 35,000 B-rep models annotated with segmentation labels. This dataset supports further exploration in B-rep-related machine learning tasks.
3. Empirical Validation: Experiments reveal BRepNet achieves superior segmentation accuracy and parameter efficiency. It outperforms graph networks and geometry-based methods like PointNet++ and MeshCNN on the Fusion 360 Gallery segmentation task by significant margins.

BRepNet's kernel configurations leverage complex topological walks to refine convolution operations, achieving a balance between performance and computational efficiency. With a demonstrated accuracy of 92.52% and IoU of 77.10%, BRepNet showcases its robustness in dealing with complex B-rep models.

Implications and Future Directions

The development of BRepNet highlights the importance of utilizing rich topological data available in B-rep models. By bypassing traditional conversion to meshes or point clouds, the architecture maintains the fidelity of CAD models, suggesting potential applications in automating complex CAD tasks, enhancing reverse engineering processes, and reconstructing parametric histories.

The introduction of the Fusion 360 Gallery segmentation dataset marks a significant step in providing researchers a robust resource for further experimentation in CAD-based machine learning research.

Future research could focus on:

• Extending BRepNet to encompass additional CAD applications, potentially improving operations like simplification, feature-based design, and manufacturability analysis.
• Investigating how alternate B-rep datasets or varied CAD features might affect the learning outcomes of the network.
• Exploring the utilization of BRepNet in conjunction with graph neural networks to enhance model expressiveness and applicability to non-standard data configurations.

Conclusion

BRepNet sets a significant precedent by directly engaging with B-rep models, embodying an advancement in how neural networks can be applied to CAD data structures. It paves the way for more nuanced applications in engineering and design, pushing the boundaries of what machine learning can achieve within the domain of CAD systems.