Integrating Expert Knowledge with Domain Adaptation for Unsupervised Fault Diagnosis

IEEE Transactions on Instrumentation and Measurement, 2021. DOI PDF

Introduction

In this repository, we provide the readers with the synthetic CWRU data set as well as the domain adaptation code. We generate the synthetic faults by injecting experts' understanding on fault patterns to the healthy signals. Details can be found in the paper.

Download Links

Dataset
- The full proposed synthetic CWRU dataset including meta data, in parquet format
- Pre-processed synthetic CWRU dataset without meta data, in numpy format
Google Colab for examples

Structure

├── data
│   ├── cwru.parquet            # CWRU dataset in parquet format, provided in the download link
│   ├── cwru_synthetic.parquet  # Synthetic CWRU dataset we generated, provided in the download link
│   ├── preprocessed            # Preprocessed data we used for our experiments
│   │   ├── XreallDEenv.npy
│   │   ├── XsynallDEenv.npy
│   │   ├── yreallDEenv.npy
│   │   └── ysynallDEenv.npy
│   └── preprocess.py           # Preprocessing code
├── code
│   └── Syn2real_CWRU.ipynb     # Notebook to reproduce our CWRU synthetic to real adaptation results. Identical to the Google Colab we provide.
└── README.md

Example code to read and preprocess the parquet files are provided in data/preprocess.py

# pip install pandas
# pip install pyarrow
import pandas as pd
df = pd.read_parquet('cwru_synthetic.parquet', engine='pyarrow')

Code

We provide a google colab research notebook for readers to better under our method. You can play with it here.

Reference

If you use our dataset or find our code helpful, please consider citing our paper:

@article{wang2021integrating,
  title={Integrating Expert Knowledge with Domain Adaptation for Unsupervised Fault Diagnosis},
  author={Wang, Qin and Taal, Cees and Fink, Olga},
  journal={IEEE Transactions on Instrumentation and Measurement},
  year={2021}
}

Acknowledgement

Links from Case Western Reserve University for the original CWRU dataset