MAML_for_Mechanical_Fault_Diagnosis

A pytorch implementation of the mechanical fault diagnosis method using Model Agnostic Meta-Learning (MAML). This code is a modification of this repository. The target of this method is one-dimensional mechanical fault signal data.

How to use

Environment

Python 3
PyTorch > 1.0
Other packages: numpy, pyyaml, tqdm, tensorboardX

Datasets

Preprocess the data into the shape of (category, number, length, 1), e.g. (10, 100, 1024, 1), and save them as npy files. The naming format is dataset_condition_split.npy, e.g. CWRU_2HP_train.npy.

Add the register module at the end of datasets\bearingdataset.py in the same way as in the file.

An example of pre-processing for CWRU data is given here.

python preprocessing.py --data_dir=$path_to_data_folder$ --out_dir=$path_to_save_folder$ --data_name=CWRU

Configurations

Add or modify configuration files in configs/.

Modifying the model

The backbone network consists of the encoder and the classifier, taking care that the shape of the output of the encoder matches the shape of the input required of the classifier.

If you add new model files, remember to update models/classifiers/__init__.py and models/encoders/__init__.py.

Training

python train.py --config=configs/cwru_ticnn_train.yaml

Testing

python test.py --config=configs/cwru_ticnn_test.yaml

Notes

You can carry out different experiments by：

using different datasets
changing different models
freezing the head or body of the backbone network
etc

If the input data is two-dimensional, you can refer to the reference to modify the code.

Reference

@misc{pytorch_maml,
  title={maml in pytorch - re-implementation and beyond},
  author={Mu, Fangzhou},
  howpublished={\url{https://github.com/fmu2/PyTorch-MAML}},
  year={2020}
}

TianyuanYang/MAML-for-Mechanical-Fault-Diagnosis