This repository contains notebooks and scripts to get you started experimenting with predictive maintenance models.
The data for this project can be found here.
This repository is meant to accompany this blog post.
To generate necessary train/test data, execute the following command:
make data
This will produce a data/
folder, which will contain the following:
data
├── processed
│ ├── RUL.csv
│ ├── test.csv
│ └── train.csv
└── raw
...
The files in processed/
are cleaned and consolidated versions of the original files (stored in raw/
), saved as CSVs.
The test.csv
and train.csv
files are in the following format:
Column | Description |
---|---|
dataset_id | id of the dataset where this instance is found |
unit_id | id of engine (unique in each dataset) |
cycle | number of operational cycles since beginning of engine operation |
setting 1 | value of operational setting 1 |
setting 2 | value of operational setting 2 |
setting 3 | value of operational setting 3 |
sensor 1 | value of sensor 1 |
... | ... |
sensor 21 | value of sensor 21 |
In the case of train.csv
, observations continue until the time of failure of the unit. In test.csv
, observations cease some number of cycles before engine failure.
Column | Description |
---|---|
dataset_id | id of the original dataset of this instance |
unit_id | id of engine (unique in each dataset) |
rul | the remaining useful life of this unit, after its maximum cycle in the test dataset |
For some example visualizations and exploration, see the example notebook in notebooks/example.ipynb
.