This is a re-implementation of FeatureNet in Tensorflow 2. FeatureNet is a deep learning architecture for machining feature recognition that utilises a voxel representation and 3D CNN.

The code is based on this original paper. This paper's original code can be found here and dataset is available here.

• Python > 3.8.5
• Tensorflow > 2.2.0
• Numpy > 1.19.1
• h5py > 1.10.6
• scipy > 1.5.2
• scikit-image > 0.17.2
• matplotlib > 3.3.2
• Binvox ^{(Opensource software)}

• To create voxel models, CAD models be converted to .stl format. These must have the following naming convertion "{class_num}-{index_num}" e.g. 0-100.stl.
• The binvox .exe must be placed in the dataset directory with the stl files. A shell script is provided in /utils to convert all the stls called convert_to_voxel.sh, this must also be placed in the directory. Below is an example of running the shell script for voxel models of resolution 64^3.

bash ./convert_to_voxel.sh 64

• An additional shell script convert_to_voxel.sh is also provided that will rotate the voxel model to create additional samples.
• The voxel models are split into training/validation/test subsets and batches are created then stored in h5df files. These h5df files are what are loaded during training and testing. This dataset split can be achieved by running create_dataset_splits.py.

• To train FeatureNet, alter the user parameters in the training.py as required and run.

• To test a trained FeatureNet model for single machining features, alter the user parameters in test.py including the saved checkpoint found in /checkpoint directory, then run the Python file.
• To segment a CAD model with multiple machining features using a trained FeatureNet model, alter the user parameters in segmentation.py. This is run on individual binvox files.

@article{featurenet2018,
  Author = {Zhibo Zhang, PrakharJaiswal, Rahul Rai},
  Journal = {Computer-Aided Design},
  Title = {FeatureNet: Machining feature recognition based on 3D Convolution Neural Network},
  Year = {2018}
}

@article{featurenettensorflow2,
  Author = {Andrew R Colligan},
  Title = {FeatureNet Tensorflow 2},
  Journal = {https://gitlab.com/qub_femg/machine-learning/featurenet-tensorflow-2},
  Year = {2021}
}
@article{Segmentation code from: https://github.com/PeizhiShi/MsvNet}

Funding was provided by DfE.