Post on 21-Jan-2021
transcript
소개• Since 2013 July: UNIST
• 2010, Ph.D. from the University of Michigan, Ann Arbor– S. M. Wu Manufacturing Research Center– The Center of Intelligent Maintenance Systems (IMS)
• 2007, M.S. from the University of Michigan, Ann Arbor
• 2005, B.S. of Electrical Engineering from Seoul National University
• 2001, B.S. of Mechanical Engineering from Seoul National University
1
iSystems Design Lab• immune engineering for self-‐sustainable system and maintenance-‐free machine design• informatics for visualization and machine health monitoring• internet of things for smart factories
2http://isystems.unist.ac.kr/
이상진단 플랫폼
3
Internet of Things & PHM
4
IoT Sensor• System
– Wi-‐fi Micro-‐controller– IMU Accelerometer– Lithium-‐ion battery
• Training Data Acquisition– Rotor Testbed
5
Image Spec
ParticlePhoton
Broadcom BCM43362 Wi-‐Fi chipSTM32F205 120Mhz ARM Cortex M3
1MB flash, 128KB RAMhttps://store.particle.io/
IMU Sensor
3 acceleration channels16-‐bit data output1 kHz Sample Rate
https://www.sparkfun.com
Rotor Testbed
RPM 1500
Fault Mode Normal Unbalance Misalignment
Sensor Position Bearing Housing
Sensor X axis accelerometer
Sample Rate 1 kHz
* Wi-fi Communication Maximum Speed : 11 MBit/s
IoT Sensor with Machine Learning• Algorithm Embedded (C++)– Feature Extraction Function– Trained classification model
• Data processing process
6
- Data Acquisition
- FFT
- RBF Kernel
- Logistic Regression
0 100 200 300 400 500 600 700 800 900 1000-0.3
-0.2
-0.1
0
0.1
0.2
0.3
0.4
Data Number
Ampl
itude
Time Signal
10 20 30 40 50 60 70 80 90 1000
0.1
0.2
0.3
0.4
0.5
0.6
0.7FFT
Frequency
Amplitude
Machinery • Feature Vector§ 1X Amplitude§ 2X Amplitude
• Probability of Machine State
Web-‐based Dashboard• Web-‐based service– Cloud Server– Various devices can access
7
Machine status in Feature Space
Probability of ClassificationFull Spectrum
Dashboard on Mobile Devices
8
PHM with IoT and Cloud Platform• Prognostic Health Management (PHM)– Short-‐term Analysis
• IoT Sensors• Local• 현재설비상태분석• 고장상태분류
– Long-‐term Analysis• Cloud Computing• Integral• 누적된정보활용을통한트랜드분석• 시계열분석및인과관계분석
• Monitoring Systems– Data Visualization
• Intuitive Information• Interactive Information
– Web-‐based Service
9
IoT Sensor
Machinery
Machine Learning-‐ Classification-‐ Pattern Recognition
State Estimation
Cloud Platform Machine Learning-‐ Time Series Analysis-‐ Probabilistic Graph Model
Data Visualization-‐ Web Service-‐ Interactive
SensorsFeature3
Short-‐term Analysis
Long-‐term Analysis Monitoring
PHM
Diagnostics
Prognostics
Data Flow
Estimation
Maintenance
Deep Learning & PHM
10
• Prognostics and Health Management (PHM) approach• Prevent/Predict system failures
Monitoring Systems
11
Time Domain Frequency Domain Time-‐frequency Orbit Analysis
• Input : 8×1 Vector• 350 orbit images are used for validation• Total misclassification for the given test set is overall 6.0%
Gaussian Discriminant Analysis (GDA)
12
True Shape
Classified
C E H 8 T
C 69 5 0 0 0
E 1 64 0 0 0
H 0 1 68 12 0
8 0 0 2 58 0
T 0 0 0 0 70
Full SpectrumOrbit Decomposition(Radius & Phase)
z a bj= +( ), ( )z zreal imag
Feature Extraction (Real and imaginary value, 8-‐dim)
GDAClassification in 8-‐dim
Artificial Neural Network
13
• Input : 8×1 Vector• 350 orbit images are used for validation• Total misclassification for the given test set is overall 8.5%
True Shape
Classified
C E H 8 T
C 68 0 0 0 0
E 0 69 6 0 0
H 0 0 54 8 0
8 2 1 10 62 3
T 0 0 0 0 67
Structure
• 1 Input Layer§ 8 neurons
• 1 Hidden Layer§ 100 neurons
• 1 Output Layer§ 5 neurons
Orbit Analysis for Rotating Machinery• Visualize shaft movement– Vibration information
• Integrated analysis possible
14
Orbit
Axes Signal2 Sensors
Normal Unbalance Misalignment Rubbing
Proposed Idea
Orbit Shape
Machine Learning
Fault Detection
15
- Signal to image - Image pattern recognition- Deep learning(Google, Facebook, …)
- Use known-fault modes
Deep Learning• Automatic discovery of the representation for classification• Abstraction from combination of non-‐linear method• Image pattern recognition problems – hand-‐written digit recognition and face recognition– Convolutional Neural Networks (CNN)
• Hierarchical structure
16
RGB = [ 0 0 0 ]1 Pixel cannot explainany information
Small area can explain context of image
Structure of Convolutional Neural Networks Key idea of Convolutional Neural Networks
Convolutional Neural Networks (CNN)• 5 Output neurons– Max pooling– Value of neuron means the degree of activation• Probability of classification
– Each neuron represent each class• 1 0 0 0 0 = Class 1• 0.3 0.9 0 0 0 = Class 2
17
0.10.10.90.20.1
é ùê úê úê úê úê úê úë û
Convolution layer and Subsampling layer
• Autonomous orbit image pattern recognition• Training and classification process
Deep Learning on Orbit Images
18
Training Data Pre-‐processing
Pre-‐processing
Deep Learning Structure
Input Data ClassificationStructure
Training Process
Classification Process
2
1
min
ˆ ( )z
T T
z b
z b-
F -
= F F F
2
1
min
ˆ ( )z
T T
z b
z b-
F -
= F F F
wTrained
Kalman Filter & PHM(모델 기반 진단)
19
Model-‐based Diagnostics• Possible only for simple systems– Analytical– Computer simulation– But, expensive
20
Model-‐based Diagnostics• Assumptions – system can be approximated as state space representation
• If system dynamics are changed (due to fault)
• Real-‐time diagnostics via estimating matrix A– From xn and yn– From A
21
1n n n
n n n
x Axy Cx
wu
+ = +
= +
: state: observation (measurement): system matrix: measurement matrix: system noise: obervation noise
n
n
n
xyACwu
1n n n
n n n
x Axy Cx
wu
+ = += +
1n n n
n n n
Ax xy Cx
wu
+ +=
¢=+
Rotating Machinery: Misalignment• Vibration measurement• Induce misalignment during operation
• Continuous • No training step required
22
Data-driven ML classification (SVM)
Kalman Filter Estimation error
Data Visualization
23
Data Visualization
• 데이터분석결과를쉽게이해할수있도록시각적으로
표현하고전달하는과정
• 빅데이터à기계가추론à정보
24
시각화 (visualization)
사람이추론 (reasoning)
도서관 데이터 시각화• 유니스트도서관도서데이터– 106,331권의책을 4단계로분류
• Tree 시각화방법사용– 구조가아래로길게나열되어한눈에보이지않음
25
도서관 데이터 시각화
26
도서관 데이터 시각화
27
대전 타슈 자전거대여 시스템 시각화
28
29
한국 수출입 시각화
시각화 적용 사례 – PCA (주성분분석)
31
1 ( )x temperature
2(
)xvibration
1x
2x
1l
2l
PCA- Eigenvalue
PCA- Eigenvector
1u2u
1x
2x
2c
1c3c
4c
- Correlated- Linear decomposition
• Linear data dimension reduction
고유치와 고유벡터
• Eigenvalues– 새로운축이원래데이터의정보를얼마나많이가지고있는지에대한지표
• Eigenvectors– 계수 c를 통해새로운축에대한각특성인자들의중요도를유추가능
32
[ ] [ ] 1 31 2 1 2
2 4
ˆ ˆ ˆ ˆc c
u u x xc cé ù
= ê úë û
2l1l
2x
1x
2u
1u
PC 1 PC 2
주성분 분석 결과
• 기기진단모니터링에사용되는 PCA의 한계점– Eigenvectors와 Eigenvalues 값에 대한 분석이부족• 선정된축과원래데이터축사이의관계성파악에중요지표
– PCA에 대한 배경지식이없다면이해가어려움
33
PCA 시각화
34
2l1l
1x
2x
1u
2u
link
신호 기반과 시각화의 연계• 신호기반진단모듈– 학습된추론엔진을통해기기고장진단
• 현장근무자의의사결정– 데이터시각화기법을통한직관적인판단
• 진단결과조합을통해정밀한진단가능
35
Data Acquisition Feature Selection
-0.02 0 0.02-0.04
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
0.04Mode1
X axis
Y ax
is
-100 0 1000
5
10
15
20
25
30
35
40FFT of Mode1
Frequency
Ampl
itude
-0.02 0 0.02
-0.04
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
0.04
Mode2
X axisY
axis
-100 0 1000
5
10
15
20
25
30
35
40FFT of Mode2
Frequency
Ampl
itude
-0.01 0 0.01
-0.03
-0.02
-0.01
0
0.01
0.02
0.03
Mode3
X axis
Y ax
is
-100 0 1000
5
10
15
20
25
30
35
40FFT of Mode3
Frequency
Ampl
itude
-0.02 0 0.02-0.05
0
0.05Mode4
X axis
Y ax
is
-100 0 1000
5
10
15
20
25
30
35
40FFT of Mode4
Frequency
Ampl
itude
시각화
최종 기기 진단
현장 근무자 결정
신호 기반
시각화 기반
warning
warning
warning
신호 기반 진단
1: 11:
1: 1
(Z | ) ( | )( | Z )(Z | )
k k k kk k
k k
P X P X ZP XP Z
-
-
=