Jupyter Notebooks
Motivation
I am focused on making it extremely easy for anyone interested in pursuing this particular topic further to be able to duplicate my work and findings and build on top of it as well as
use as a sample for other work similar but not necesarily related or simply just learn how some of this process works hence from day one when I chose this topic I decided I would make all of it
public, open-source, free to use, copy, redistribute (that's why I chose the WTFPL license).
You will find below a list with my jupyter notebooks which were used for this project. They are a bit rough as I have not focused on making them pretty and my comments might also be a bit unprofessional but nevertheless should work as a source of inspiration.
Notebooks - Initial data preprocessing and feature selection
- step1 - populate table with list of failed drives
- step2 - populate table with data for the failed drives
- step3 - correlation between columns attributes
- step4 - more feature cleanup
- step5 - identify which features match to which vendor model
- step6 - correlation between columns attributes after more feature selection
Experiments
- RF 14 days before failure as prefailure state
- RF 30 days before failure as prefailure state
- RF 60 days before failure as prefailure state
- RF 90 days before failure as prefailure state
- BiLSTM predict RUL using 30 x 15 lookback
- BiLSTM predict RUL using 30 x 60 lookback
- BiLSTM predict RUL using 30 x 90 lookback