Skip to content

Femto

FEMTO-ST bearing dataset used in the IEEE PHM 2012 Data Challenge for RUL (remaining useful lifetime) estimation.

Description

This dataset consists of run-to-failure experiments carried on the PRONOSTIA platform. Data provided by this platform corresponds to normally degraded bearings, which means that the defects are not initially initiated on the bearings and that each degraded bearing contains almost all the types of defects (balls, rings and cage).

Data are acquired under three operating conditions (rotating speed and load force):

  • Condition 1. 1800 rpm and 4000 N: folders Bearing1_x
  • Condition 2. 1650 rpm and 4200 N: folders Bearing2_x
  • Condition 3. 1500 rpm and 5000 N: folders Bearing3_x

In order to avoid propagation of damages to the whole test bed (and for security reasons), all tests were stopped when the amplitude of the vibration signal overpassed 20g which is used as definition of the RUL.

Provided data contains the records of two acclerometers and one temperature sensor, and are splitted into the learning set of 6 experiments and the test set (truncated + full) of 11 experiments. The goal is to estimate the RUL on the test set. The learning set was quite small while the spread of the life duration of all bearings was very wide (from 1h to 7h).

Actual RULs (in second) of Test set

  • Bearing1_3: 5730
  • Bearing1_4: 339
  • Bearing1_5: 1610
  • Bearing1_6: 1460
  • Bearing1_7: 7570
  • Bearing2_3: 7530
  • Bearing2_4: 1390
  • Bearing2_5: 3090
  • Bearing2_6: 1290
  • Bearing2_7: 580
  • Bearing3_3: 820

Homepage

https://ti.arc.nasa.gov/tech/dash/groups/pcoe/prognostic-data-repository/#femto

http://www.femto-st.fr/

https://github.com/Lucky-Loek/ieee-phm-2012-data-challenge-dataset

Further Information

Monitoring data of the 11 test bearings were truncated so that participants were supposed to predict the remaining life, and thereby perform RUL estimates. Also, no assumption on the type of failure to be occurred was given (nothing known about the nature and the origin of the degradation: balls, inner or outer races, cage...)

Original Dataset

  • Type of experiments: run-to-failure.
  • Format: CSV files
  • Channels: not fixed, up to 3 (2 vibrations and 1 temperature)
    • Vibration signals (horizontal and vertical)
      • Sampling frequency: 25.6 kHz
      • Recordings: 2560 (i.e. 1/10 s) are recorded each 10 seconds
    • Temperature signals
      • Sampling frequency: 10 Hz
      • Recordings: 600 samples are recorded each minute
  • Split: ['Learning_set', 'Test_set', 'Full_Test_Set']
  • Label: None
  • Size: ~ 3 Gb, unzipped

Download

https://github.com/Lucky-Loek/ieee-phm-2012-data-challenge-dataset

Built Dataset

Split: ['train', 'test', 'full_test']

Features

  • 'signal': {'vibration', 'temperature'}
  • 'sampling_rate': { 'vibration': 25600, 'temperature': 10 },
  • 'metadata':
    • 'ID': ['Bearing1','Bearing2','Bearing3']
    • 'OriginalSplit': ['Learning_set', 'Test_set', 'Full_Test_Set']
    • 'RotatingSpeed'
    • 'LoadForce'

Notes

  • The split 'test' is a subset of the split 'full_test'.
  • When both being recorded, 'vibration' and 'temperature' correspond to the same experiment and should be understood as two channels. However, being sampled at different sampling rate they cannot be extracted simultaneously in form of a 3-channels signal.
  • The original timestamp of sampling is not regular. We made the choice to not include the timestamp data.
  • RUL cannot be extracted due to a strange error, see the comments in the source code.

Tutorial