Upload
others
View
4
Download
0
Embed Size (px)
Citation preview
ELIS – Multimedia Lab ELIS – Multimedia Lab
Infrared and Vibration based Bearing Fault Detection Using Neural Networks
Olivier Janssens, Lothar Verledens, Raiko Schulz ,Veerle Ongenae
Kurt Stockman, Mia Loccufier, Rik Van de Walle, Sofie Van Hoecke
AITA - 2015 - PISA
1/21
ELIS – Multimedia Lab ELIS – Multimedia Lab
Overview
• Bearings and bearing fault causes
• Experiments and data set creation
• Fault-detection architecture
• Results
• Conclusion
2/21
ELIS – Multimedia Lab ELIS – Multimedia Lab
Bearings: what they do
3/21 Bearings & faults
Experiment & data set
Architecture Results &
conclusion
ELIS – Multimedia Lab ELIS – Multimedia Lab
Bearings: example where they are used
4/21 Bearings & faults
Experiment & data set
Architecture Results &
conclusion
ELIS – Multimedia Lab ELIS – Multimedia Lab
Bearing failure
Downtime
Costs
Fault propagation
Reduced lifetime
5/21 Bearings & faults
Experiment & data set
Architecture Results &
conclusion
ELIS – Multimedia Lab ELIS – Multimedia Lab
Bearing failure
Improper lubrication
Delayed lubrication renewal
Inadequate lubrication
Lubricant contamination
Unsuitable lubrication
Inadequate bearing
selection
Improper mounting
Load imbalance
Misalignment
Indirect failures
Manufacturing errors
10% 5% 1% 4%
25 %
15%
20%
20 %
[1] C. Radu, The most common causes of Bearing Failure and Importance of Bearing Lubrication,
RKB Technical review, 2010
6/21 Bearings & faults
Experiment & data set
Architecture Results &
conclusion
ELIS – Multimedia Lab ELIS – Multimedia Lab
Goal
Detect and identify which fault(s)/ condition(s) is/are
present using thermal imaging and vibration data
7/21 Bearings & faults
Experiment & data set
Architecture Results &
conclusion
ELIS – Multimedia Lab ELIS – Multimedia Lab
Setup & faults/conditions
1. Servo-motor 5. Disk 9. Metal plate
2. Coupling 6. Shaft 10. Field of view
3. Bearing housing 7. Accelerometer 11. Thermal camera
4. Bearing 8. Thermocouple
8/21 Bearings & faults
Experiment & data set
Architecture Results &
conclusion
ELIS – Multimedia Lab ELIS – Multimedia Lab
No fault (NF)
Mildly reduced lubrication (MRL)
Heavily reduced lubrication (HRL)
Outer raceway fault (ORF)
Setup & faults/conditions
9/21 Bearings & faults
Experiment & data set
Architecture Results &
conclusion
ELIS – Multimedia Lab ELIS – Multimedia Lab
1 2 3 4 5
6 7 8 9 10
11 12 13 14 15
16 17 18 19 20
21 22 23 24 25
26 27 28 29 30
31 32 33 34 35
36 37 38 39 40
NF
NF-IM
MRL
MRL-IM
HRL
HRL-IM
ORF
ORF-IM
Be
arin
g 1
Be
arin
g 2
Be
arin
g 3
Be
arin
g 4
Be
arin
g 5
Data:
40 recordings (5 bearings * 8 faults/conditions)
1 hour video per recording
10 minute vibrations per recording
Corrected for ambient temperature.
10/21 Bearings & faults
Experiment & data set
Architecture Results &
conclusion
ELIS – Multimedia Lab ELIS – Multimedia Lab
Multi-sensor solution:
Thermal images - preprocessing:
Region of interest detection using Gaussian mixture models [2]
[2] Z. Zivkovic. Improved adaptive gaussian mixture model for background subtraction. ICPR 2004. pages
28–31 Vol.2, Aug 2004.
11/21 Bearings & faults
Experiment & data set
Architecture Results &
conclusion
ELIS – Multimedia Lab ELIS – Multimedia Lab
Bearings & faults
Experiment & data set
Architecture Results &
conclusion
Multi-sensor solution:
Thermal images - preprocessing:
Use ROI as timeseries
12/21
ELIS – Multimedia Lab ELIS – Multimedia Lab
Bearings & faults
Experiment & data set
Architecture Results &
conclusion
Multi-sensor solution:
Thermal images - preprocessing:
Use ROI as timeseries
13/21
ELIS – Multimedia Lab ELIS – Multimedia Lab
Multi-sensor solution:
Thermal images – dataformat:
Use ROI as timeseries
It(x,y) It+1(x,y) … I225(x,y)
1 hour
• 500 pixels from the seal
• 1000 pixels from the top
of the housing
14/21 Bearings & faults
Experiment & data set
Architecture Results &
conclusion
ELIS – Multimedia Lab ELIS – Multimedia Lab
Fault detection architecture
MRL
MRL – IM
ORF
ORF – IM
HB
HB – IM
HRL
HRL – IM
ORF
ORF – IM
HB
HB – IM
HRL
HRL – IM
MRL
MRL – IM
HB
HB – IM
HRL
HRL – IM
ORF
ORF – IM
HRL
HRL – IM
HB
HB – IM
MRL
MRL – IM
ORF
ORF – IM
HB
HB – IM
HRL
HRL – IM
MRL: Mildly reduced lubrication
ORF: Outer raceway fault
HB: Healthy bearing
HRL: Heavily reduced smearing
Neural network
Random forest classifier
Support vector machine
15/21 Bearings & faults
Experiment & data set
Architecture Results &
conclusion
ELIS – Multimedia Lab ELIS – Multimedia Lab
Multi-sensor solution – Neural network 1:
Timeserie seal
Timeserie housing top
Timeserie housing top 2
255 n
odes
255 n
odes
255 n
odes
325 n
odes
40
nodes
MRL (-IM)
HRL (-IM)
ORF(-IM)
HB(-IM)
Architecture: 675 325 40 2
16/21 Bearings & faults
Experiment & data set
Architecture Results &
conclusion
ELIS – Multimedia Lab ELIS – Multimedia Lab
Multi-sensor solution – Neural network 2:
Timeserie housing top 1
255 n
odes
100 n
odes
HB (-IM)
HRL (-IM)
Architecture: 255 100 2
17/21 Bearings & faults
Experiment & data set
Architecture Results &
conclusion
ELIS – Multimedia Lab ELIS – Multimedia Lab
Neural network techniques used:
- Training: Stochastic gradient descent + momentum with Backpropagation
- Activation function hidden nodes: Rectified linear units
- Activation output nodes: Softmax
- Loss function: Cross entropy
18/21 Bearings & faults
Experiment & data set
Architecture Results &
conclusion
ELIS – Multimedia Lab ELIS – Multimedia Lab
Vibration part:
Vibration data
Vibration data
Peak to peak features
Kurtosis features Random forest
ORF (-im)
HRL (-IM)
HB(-IM)
Amplitude at the rotation
frequency (25 Hz) Linear SVM
Imbalance
Balance
Data Features Classification
model Classes
19/21 Bearings & faults
Experiment & data set
Architecture Results &
conclusion
ELIS – Multimedia Lab ELIS – Multimedia Lab
Results
Fault IR-based precision Multi-sensor precision
HB 50 % 100 %
HB - IM 70 % 100 %
MRL 100 % 100 %
MRL - IM 80 % 80 %
HRL 70 % 100 %
HRL - IM 70 % 90 %
ORF 30 % 100 %
ORF - IM 40 % 100 %
Average 63,75 % 96,25 %
20/21 Bearings & faults
Experiment & data set
Architecture Results &
conclusion
ELIS – Multimedia Lab ELIS – Multimedia Lab
Acknowledgment
21/21
This work was partly funded by the O&M Excellence project, a
VIS project of the Institute for the Promotion of Innovation through
Science and Technology in Flanders (IWT), and has been
performed in the framework of the Offshore Wind Infrastructure
Application Lab (http://www.owi-lab.be).
ELIS – Multimedia Lab ELIS – Multimedia Lab
Thank you for listening !
Questions ?
22/21