Anomaly Detection with Autoencoder

Inspired by Vegard Flovik's post on Medium
Data from here

Purpose:

We try to detect anomalies from "IMS Bearing Data". We used "Test1" data which has 8 channels;

Bearing 1 – Ch 1&2; Bearing 2 – Ch 3&4;
Bearing 3 – Ch 5&6; Bearing 4 – Ch 7&8.

Data recorded every 10 minutes (except the first 43 files were taken every 5 minutes).
At the end of the test-to-failure experiment, inner race defect occurred inbearing 3 and roller element defect in bearing 4.

Steps:

1. prepare data
2. visualize data
3. build autoencoder ANN
4. get results
5. ...
6. profit?

How to run?

1. Create "data" folder.
2. Download data from here.
3. unzip "1st_test" folder to "data" folder we created.
4. run "anomaly_detection.ipynb" bu jupyter notebook.

Why I didn't automated downloading and unzipping? Because I was too lazy!

What to do next?

1. Improve training accuracy (maybe after many steps, we can reach some kind of overfit which might be good for this task)