![Page 1: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/1.jpg)
DEEP NEURAL NETWORK FOR PROGNOSTICS
AND HEALTH MANAGEMENT
Zhe Yang
![Page 2: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/2.jpg)
2
k
j l
An Artificial Neural Network (ANN) is composed by several simple computational units (also called
nodes or neurons) directionally connected by weighted connections organized in a proper architecture
Input layer Hidden layer Output layer
Neuron
(node)
Bias π€ππ π€ππ
0
3
1
jk j
h h
j k
k
wf x wu=
= +
1
0
2o o h
l j
j
lj lw wu f u=
= +
Activate function
Artificial Neural Network
π¦πβ π¦π
π
π’πβ π’π
π
![Page 3: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/3.jpg)
3
ANN-based PHM
Feature
extraction(Variance/Maximum
/Minimum/Skewnes
s/Kurtosis/Peak/Wa
veletβ¦)
β¦
f1
f2
f3
fN
fsel1
fsel2
fsel3
Feature
selection(Lasso/
Filter/wrapper
methodβ¦)
Raw signal
Statistical Indicators
Expert intervention!
ANN
β’ Abnormal
condition
β’ Fault class
β’ RUL
![Page 4: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/4.jpg)
4
Limitations of ANN
β’ Expert intervention for feature extraction and selection
β’ ANN UPPER LIMIT of learning capability
β’ MNIST database of handwritten digits [1]
Training set: ππ =60,000 samples
Test set: 10,000 samples
0
1
2
3
4
5
6
7
8
9
What are they?
Accuracy
ANN [2] 99.3
Deep Neural Network[3] 99.65
[1] http://yann.lecun.com/exdb/mnist/
[2] Simard P Y, Steinkraus D, Platt J C. Best practices for convolutional neural networks applied to visual document analysis[C]//null. IEEE, 2003: 958.
[3] Ciresan et al. Neural Computation 10, 2010 and arXiv 1003.0358, 2010
π₯11
π₯21
π 1π 2
π₯12
π₯22
π ππ
π₯1ππ
π₯2ππ
β¦
β¦
β¦β¦β¦
![Page 5: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/5.jpg)
5
Neural Network-based PHM
Feature
extraction(Variance/Maximum
/Minimum/Skewnes
s/Kurtosis/Peak/Wa
veletβ¦)
β¦
f2
f3
fN
fsel1
fsel2
fsel3
Feature
selection(Lasso/
Filter/wrapper
methodβ¦)
TaskRaw signal
Raw signal Deep Neural Network Task
Extract & select useful features automatically!
ANN
![Page 6: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/6.jpg)
6
Shallow and Deep Neural Network
architecture
Shallow Neural Network Deep Neural Network (DNN)
Number of hidden layers: typically <3 Number of hidden layers: >=3
Why DNN was rarely used ten years agoοΌ
![Page 7: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/7.jpg)
7
How to train a DNN?
β’ Error back propagation:
Input layer
Hidden
layers
Output layer
πΈ =1
2ππ
π=1
ππ
π’ππ β π‘π
2
TRUE l-th output
ANN l-th output
Number of output
neuronsππΈ
ππ’ππ =
π’ππ β π‘πππ
π€ππ4(π+1)
= π€ππ4(π)
β πππΈ
ππ’ππ
ππ’ππ
ππ¦ππ
ππ¦ππ
ππ€ππ4
π€π4π3(π+1)
= π€π4π3(π)
β πππΈ
ππ’ππ
ππ’ππ
ππ¦ππ
ππ¦ππ
ππ’π4β
ππ’π4β
ππ¦π3β
ππ¦π3β
ππ€π4π3
π€π3π2(π+1)
= π€π3π2(π)
β πππΈ
ππ’ππ
ππ’ππ
ππ¦ππ
ππ¦ππ
ππ’π4β
ππ’π4β
ππ¦π3β
ππ¦π3β
ππ’π2β
ππ’π2β
ππ¦π2β
ππ¦π2β
ππ€π3π2
π€π2π1(π+1)
= π€π2π1(π)
β 0
π€π1π(π+1)
= π€π1π(π)
β 0
Problem: fail to
tune weights!
Product of 3 partial derivatives
Product of 5 partial derivatives
Product of 7 partial
derivatives
Partial derivatives are small numbers (<1),
the product of many partial derivatives is β0
Gradient
vanishing
π€ π+1 = π€ π β πππΈ
ππ€
π’ππ
![Page 8: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/8.jpg)
8
How to train a DNN: Layer-wise training
β’ Pre-training
Breakthrough: 2006 [1]
[1] Hinton G E, Salakhutdinov R R. Reducing the dimensionality of
data with neural networks[J]. science, 2006, 313(5786): 504-507.
β’ Stacking
![Page 9: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/9.jpg)
9
DNN pre-training β Autoencoder
Input layer
Encoder
Decoder
Hidden layer
Output layer
Same number
of neurons
Input patterns
Reconstructed input patterns
Similar information
content
![Page 10: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/10.jpg)
10
Sparse autoencoder: training
Encoder Decoder
Input samplesReconstructed
input samples
( ) ( )2 2
1
1 1Λ * *
2
pn
p p
sparse
pp
E Rn
=
= β + + Wx x
β’ Cost function
Reconstruction
Error
Sparse
decomposition
of the input
Constrain the
value of inner
weights
π π , π = 1,β¦ , ππ ΰ·π π , π = 1,β¦ , ππ
![Page 11: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/11.jpg)
11
Sparse autoencoder: sparsity
β’ Sparsity
Input samples
π π , π = 1,β¦ , ππ
β¦
0 0 0 0 0
β¦
β¦
0 0 0 0 0
β¦
1 1
1 hidden neuron only responds to 1 kind of pattern
Sparse
representation of
input
![Page 12: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/12.jpg)
12
Sparse autoencoder: sparsity
j
ΰ·ππ =1
ππ
π=1
ππ
πβ
π=1
4
π₯πππ€ππ +π€π0
π₯11
π₯21
π₯31
π₯41
π 1π 2
π₯12
π₯22
π₯32
π₯42
π ππ
π₯1ππ
π₯2ππ
π₯3ππ
π₯4ππ
β¦
β¦
Output of hidden
neuron j given input π πAverage output of hidden neuron
j over the training set
πβ π 1 =0,πβ π 2 =1, πβ π ππ =0β¦
ΰ·ππ =π
ππ Very small number when π βͺ ππ
β’ Let us consider hidden neuron j
k
π 2
β¦
j
β¦
β¦
Input samples
(many different patterns)
π = 3, the number of
![Page 13: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/13.jpg)
13
Sparse autoencoder: sparsity
j
β’ To obtain the sparse decomposition of input samples, we set an expected average output for
hidden neurons, π.
ΰ·ππ =1
ππ
π=1
ππ
ππ π π
π = 0.3
π
Training objective for sparsity
KL divergence: Difference between ΰ·ππ and π
( ) ( )2 2
1
1 1Λ * *
2
pn
p p
sparse
pp
E Rn
=
= β + + x x W
π
πβ
KL παΰ·ππ =
π
πβ
π logπ
ΰ·ππ+ 1 β π log
1 β π
1 β ΰ·ππ
ΰ·ππ
ΰ·ππ = π
KL παΰ·ππMinimize
![Page 14: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/14.jpg)
14
Build a DNN using autoencoders
2000
2000
1000
300
1000
1000
300
300
10
2000
1000
300
10
Objective
100
100
100
100
Autoencoder 1
Autoencoder 2
Autoencoder 3
Autoencoder 4
Step 1. train
autoencoders
Input patterns
Target
Input patterns
Classification/
regression
![Page 15: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/15.jpg)
15
Build a DNN using autoencoders
2000
2000
1000
300
1000
1000
300
300
10
100
100
100Step 1
2000
1000
300
10
300
1000
2000
100
100
Step 2. Stacking and
fine-tuning
2000
1000
300
10
100
Step 3. Throw away
βDecoderβ and
supervised training
Classification/
regression
Input patterns Input patterns Input patterns
Target
![Page 16: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/16.jpg)
16
DNN: Difficulty to decide the number of neurons
20
00
10
00
300
10
10
0
Cla
ssific
atio
n/
regre
ssio
n
How to decide the
numbers of layers and
neurons?
Sparse autoencoder: extract interesting features in
the data relatively independent from the layers and
neurons
β¦
0 0 0 0 0
β¦
1
β¦
β¦
0 0 01
![Page 17: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/17.jpg)
17
Comparison: Shallow and Deep Neural
Network
Shallow Neural Network Deep Neural Network (DNN)
Number of hidden layers: typically <3 Number of hidden layers: >=3
Training Training
Accuracy Accuracy
Big data Big data
Back propagation
β
β
Γ
Γ
Raw data Γ Raw data β
Layer-wise training using
autoencoders
![Page 18: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/18.jpg)
18
Case study
DNN for bearing fault detection
![Page 19: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/19.jpg)
19
Case study
β’ [1]β’ Available information: run-to-failure
vibration signal of 4 bearings
[1] Benchmark available at Nasa prognostic repository (https://ti.arc.nasa.gov/tech/dash/groups/pcoe/prognostic-data-repository/). J. Lee, H.
Qiu, G. Yu, J. Lin, and Rexnord Technical Services (2007). IMS, University of Cincinnati. "Bearing Data Set", NASA Ames Prognostics Data
Repository (http://ti.arc.nasa.gov/project/prognostic-data-repository), NASA Ames Research Center, Moffett Field, CA
State at the end of
experiment
Healthy
Healthy
Failed
(Inner race)
Failed
(Roller element)
B1
B2
B3
B4When?
![Page 20: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/20.jpg)
20
Case study- data collection
Sample
Acc
eler
atio
n m
/s2
time
Snapshot (1s) snapshot snapshot
β¦β¦
B3
10 min
Failed
(Inner race)
![Page 21: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/21.jpg)
21
Case study: data preprocessing
time
Snapshot (1s) snapshot snapshotB3
10 min
20480 samples
Samples
Coeffic
ients
Mean
Co
effic
ients
Continuous Wavelet Transform
![Page 22: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/22.jpg)
22
Case study: train sparse autoencoder
Extracted features
1333
700
5
700
1500
1333
200
200
1500
β¦
B3
Run-to-failure
trajectory
![Page 23: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/23.jpg)
23
Degradation
trend
Abrupt jump
at the end
π =
π β 1
π0
π + 1
π
If π > 0
If π = 0
If π < 0
Mann-Kendall
metric
π =
π=1
πβ1
π=π+1
π
π ππ ππ β ππ , π =π π β 1 2π + 5
18
Feature
Test of signal
monotonicity
(Mann-Kendall)
1 27.6913
2 -21.7529
3 9.6151
4 -13.6292
5 -11.1801
Zoom of Extracted features
Case study: feature evaluation
![Page 24: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/24.jpg)
24
Case study: train sparse autoencoder
Extracted feature 1
Degradation trend
![Page 25: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/25.jpg)
25
Case study: training of DNN
1333
700
5
700
1500
1333
200
200
15001333
700
5
200
1500
Fault detection(classification layer)
![Page 26: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/26.jpg)
26
Case study: training of DNN
1333
700
5
200
1500
Fault detection(classification layer)
Snapshot Label Health State
1~1616 0 Healthy
1617~2027 0.5 Possibly Failed
2028~2156 1 Failed
β’ Training set
B3
Snapshot 1 Snapshot 2 Snapshot 3
Onset of failure: 1617 [1], 2027 [2]
[1] Qiu, H., J. Lee, J. Lin, and G. Yu, Wavelet filter-based weak signature detection method and its application on rolling element bearing prognostics. Journal of sound and vibration, 2006. 289(4): p. 1066-1090.
[2] Hasani, R.M., G. Wang, and R. Grosu, An Automated Auto-encoder Correlation-based Health-Monitoring and Prognostic Method for Machine Bearings. arXivpreprint arXiv:1703.06272, 2017.
Output: health state
Failed
(Inner race)
![Page 27: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/27.jpg)
27
1333
5
1500
700
200
β’ Developed model
Fault detection(classification layer)
Healthy
Healthy
De
gra
da
tio
n le
ve
lD
eg
rad
atio
n le
ve
l
Case study: results
B1
B2
Robust when the signals of B1 and B2 are
influenced by failed bearings
![Page 28: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/28.jpg)
28
Healthy
Failed
Snapshot Label Health State
1~1616 0 Healthy
1617~1760 0.5 Possibly Failed
1761~2156 1 Failed
β’ Result
The onset is identified at snapshot 1680
with threshold Th=0.5, which lies in the
possibly failed range [1617, 1760]
[1] Qiu, H., J. Lee, J. Lin, and G. Yu, Wavelet filter-based weak signature detection method and its application on rolling element bearing prognostics. Journal of sound and vibration, 2006. 289(4): p. 1066-1090.
[2] Hasani, R.M., G. Wang, and R. Grosu, An Automated Auto-encoder Correlation-based Health-Monitoring and Prognostic Method for Machine Bearings. arXiv preprint arXiv:1703.06272, 2017.
[3] Yu, J., Health condition monitoring of machines based on hidden Markov model and contribution analysis. IEEE Transactions on Instrumentation and Measurement, 2012. 61(8): p. 2200-2211.
Case study: results
B4
De
gra
da
tio
n le
ve
l
Failed
(Roller element)
Detect multiple failure
![Page 29: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/29.jpg)
29
Conclusions
β’ Shallow and deep neural network
β’ Training of DNN: autoencoders
β¦
β’ Sparse autoencoder
β’ Case study
KL παΰ·ππ
VS.
![Page 30: DEEP NEURAL NETWORK FOR PROGNOSTICS AND ......method and its application on rolling element bearing prognostics.Journal of sound and vibration, 2006. 289(4): p. 1066-1090. [2] Hasani,](https://reader035.vdocuments.site/reader035/viewer/2022071107/5fe12f8690850a32a812ea88/html5/thumbnails/30.jpg)
30