Residual Error based Anomaly Detection Using Auto-Encoder in SMD Machine Sound
About
This is the implementation of the paper "Residual Error based Anomaly Detection Using Auto-Encoder in SMD Machine Sound" by Dong Yul Oh and Il Dong Yun.
For more information check out the paper [PDF] on MDPI Sensors.
All of these scripts are based on the [sample] dataset provided in our paper. Note that all rights reserved to [Crevis Co., Ltd], and only a few samples for the SMD machine sound can be released. The experimental results of this demo may differ (have poorer performance) from those in the paper because of the lack of training data.
Getting started
Dependencies
Python 3.5.0
Tensorflow 1.4.0
Psutil 5.4.1
Numpy 1.13.3
Matplotlib 1.2.0
Librosa 0.5.1
Training & Validation
Script:
setup_dataset.py: data pre-processing and building tfrecords.
model.py: the network architecture as proposed in our paper.
main.py: training script for this model.
Usage:
python setup_dataset.py transforms the audio files in each class into the tfrecords using STFT.
python setup_dataset.py test checks if the tfrecords is created properly.
python main.py --train=true trains the model only with normal data.
python main.py measures AUC and the residual errors for each abnormal and normal test case.
BibTeX
If you use this code in your project, please cite our paper:
@Article{s18051308,
AUTHOR = {Oh, Dong Yul and Yun, Il Dong},
TITLE = {Residual Error Based Anomaly Detection Using Auto-Encoder in SMD Machine Sound},
JOURNAL = {Sensors},
VOLUME = {18},
YEAR = {2018},
NUMBER = {5},
ARTICLE NUMBER = {1308},
URL = {http://www.mdpi.com/1424-8220/18/5/1308},
ISSN = {1424-8220},
DOI = {10.3390/s18051308}
}