Code for paper "FilletRec: A Lightweight Graph Neural Network with Intrinsic Features for Automated Fillet Recognition".
This work builds upon several open-source implementations and research efforts. We sincerely thank the authors for their valuable contributions to the community.
- Project Code: Modified from Hierarchical-CADNet
- Dataset Creation: Modified from code in
datasetof AAGNet - Methods' Comparison: The author of Defillet has helped us run several cases in our paper
Automated recognition and simplification of fillet features in CAD models is critical for CAE analysis, yet it remains an open challenge. Traditional rule-based methods lack robustness, while existing deep learning models suffer from poor generalization and accuracy on complex fillets due to their generic design and inadequate training data. To address these issues, this paper proposes an end-to-end, data-driven framework specifically for fillet features. We first construct and release a large-scale, diverse benchmark dataset for fillet recognition to address the inadequacy of existing data. Based on it, we propose FilletRec, a lightweight graph neural network. The core innovation of this network is its use of pose-invariant intrinsic geometric features, such as curvature, enabling it to learn more fundamental geometric patterns and thereby achieve high-precision recognition on complex geometric topologies. Experiments show that FilletRec surpasses state-of-the-art methods in both accuracy and generalization, while using only 0.2%-5.4% of the parameters of baseline models, demonstrating remarkable model efficiency. Finally, the framework completes the automated workflow from recognition to simplification by integrating a novel geometric simplification algorithm.
- python >= 3.8
- tensorflow >= 2.13.0
- pythonocc-core >= 7.5.1 (more info here: https://github.com/tpaviot/pythonocc-core.git)
- occwl (more info here: https://github.com/AutodeskAILab/occwl.git)
- numpy
- scikit-learn
git clone https://github.com/Miss-Hedgehog/FilletRec.git
cd FilletRec
conda env create -f environment.yml
conda activate filletrec
If the environment.yml installation fails, please install the required packages manually.
Our synthetic Fillet datasets have been publicly available on Kaggle. Please read About Dataset on Kaggle to learn more information about our dataset structure.
For fillet recognition, set train_set_path and val_set_path in train.py, this network can be trained using:
python ./train.py
The best checkpoint based on the smallest validation loss will be stored in a folder called checkpoint, its filename is based on the network's parameter setup and timestamp with the format of .weights.h5.
For testing on test dataset, set checkpoint_path and test_set_path in test.py, this network can be tested using:
python ./test.py
To test a single CAD model with a trained network model and save the predicted labels in a *.step file to the save directory, please specify step_dir, save_dir, step_name, checkpoint_path in test_and_save.py, then run:
python ./test_and_save.py
A directory of *.step files can be viewed using the visualizer.py script, in which each label has a unique color. Set dataset_dir, then run:
python ./visualizer.py
We propose a three-stage core strategy Extend-Intersect-Clean to remove recognized fillets in CAD models.
To extend the boundary mesh, set the filename and step_file parameters, where the step_file should alredy contain the necessary fillet face labels, then run:
python ./simplify_fillet_mesh.py
In simplify_fillet_with_mesh() function, the offset_dis parameter controls the mesh extension distance. Based on your specific model, you can check the extended mesh file *_sim.vtk to adjust this value. Also, you can adjust the conformality of the mesh by resetting defletion parameter.
Sometimes, setting a uniform offset_dis for all surfaces may not be suitable, you can modify code of add_mesh_at_boundary() function to extend triangle mesh on different surfaces by different distances to accomodate complex fillet types. For example:
......
if face_count == face_num:
offset_distance = distance1
if face_count in [face_num1, face_num2, ... ,face_numx]:
offset_distance = distance2
......
The mesh intersection operation in our framework is implemented based on the robust algorithm proposed by by Liu's Robust and fast local repair for intersecting triangle meshes, read this paper for more information about mesh intersection.
To remove non-manifold structures and isolated components resulting from the intersection, set filename and vtk_file has be boolean intersected in delete_boundary_tri.py, then run:
python ./delete_boundary_tri.py
After doing all the command mentioned above, the vtk file named *_result.vtk is the defilleted result of the CAD model.