Upload
dangcong
View
216
Download
1
Embed Size (px)
Citation preview
System Reliability &Safety LaboratorySRSL 1
Chao Hu
Assistant Professor
Department of Mechanical Engineering
Iowa State University
Toward Reliable Engineered System Design:RELIABILITY-BASED DESIGN AND PROGNOSTICS AND HEALTH MANAGEMENT (PHM)
System Reliability &Safety LaboratorySRSL 2
Motivation
I-35W Bridge collapse due to faulty design, Aug. 2007
Consequence: 13 deaths, 145 injured, $60 million loss
Wind turbine collapse due to faulty maintenance, Feb. 2008
Consequence: Collapse of whole wind turbine
Boeing 787 Dreamliner fire due to overheated Li-ion battery
Japan Airlines (JAL) in Boston, Jan. 2013
Consequence: Over $1.1 million daily loss due to groundings
System Reliability &Safety LaboratorySRSL 3
Power transformer fire due to faulty bushing, Jul. 2002
Consequence: $5 million property & business loss
UPS flight fire possibly due to overheated Li-ion battery, Feb. 2006
Consequence: 3 injured, loss of whole airplane
Research Questions:
Q1. Is it possible to design a system with near-zero failure probability?
Q2. Is it possible to anticipate and prevent failures during system operation?
Motivation
System Reliability &Safety LaboratorySRSL 4
Reliability-
Based Design
Fatigue Life
PD
F
Optimized
Initial
Limit
Scientist @ MDT / Faculty @ ISU
Research Timeline
PhD Student @ UMD
2007 2009 2011 2013
Prognostics and Health Management (PHM)
Sensing Reasoning Prognostics
Remaining Useful Life (RUL)
PD
F
Health Index 1
He
alt
h I
nd
ex
2
Battery Prognostics
Deep Brain StimulatorsLi-Ion Rechargeable
2015
System Reliability &Safety LaboratorySRSL 5
Scientist @ MDT / Faculty @ ISU
Research Timeline
PhD Student @ UMD
2007 2009 2011 2013
Photo credit
http://www.mpoweruk.com
Metal
particle
(−) Electrode
Separator
(+) Electrode
Inside a Lithium-Ion
Battery
2015
System Reliability &Safety LaboratorySRSL 6
Design of Control Arm (US Army): Methodology
0
[Life at 1st hotspot=Limit]
Failure Surface G1 = 0
X1: − thickness of
comp. 1
[Lives at both hotspots > Limit]
Reliable Region
Failure Surface G2 = 0
[Life at 2nd hotspot=Limit]
Deterministic
Optimum
1st hotspot
2nd hotspot
X2: −
th
ick
ne
ss o
f
com
p.
2
Reliability-Based Design
Initial
Design
System Reliability &Safety LaboratorySRSL 7
Design of Control Arm (US Army): Optimization Results
Initial
31.473
0.3235
1.0000
1.0000
1.0000
0.0050
1.0000
Optimum
32.717
0.9987
1.0000
1.0000
1.0000
0.9989
0.9991
Weight
R6
R43
R46
R80
R87
R42
Reliability
0 199.87%
Reliability-Based Design
X3 X4
X5
X7
X6
X2
X1
System Reliability &Safety LaboratorySRSL 8
Design of Control Arm (US Army): Optimization Results
Initial
31.473
0.3235
1.0000
1.0000
1.0000
0.0050
1.0000
Optimum
32.717
0.9987
1.0000
1.0000
1.0000
0.9989
0.9991
Weight
R6
R43
R46
R80
R87
R42
Reliability
0 199.87%
Reliability-Based Design
Initial Stress
Contour
Final Stress
Contour
Optimum design
Initial design
System Reliability &Safety LaboratorySRSL 9
Research Timeline
Reliability-
Based Design
Fatigue Life
PD
F
Optimized
Initial
Limit
Prognostics and Health Management (PHM)
Sensing Reasoning Prognostics
Remaining Useful Life (RUL)
PD
F
Health Index 1
He
alt
h I
nd
ex
2
Battery Prognostics
Deep Brain StimulatorsLi-Ion Rechargeable
Scientist @ MDT / Faculty @ ISUPhD Student @ UMD
2007 2009 2011 2013 2015
System Reliability &Safety LaboratorySRSL 10
Human Life-time
He
alt
h c
on
dit
ion
Human PHM Process
Prognostics and Health Management (PHM)
Perfectly healthy
Death limit
Medical treatment
System Reliability &Safety LaboratorySRSL 11
Prognostics and Health Management (PHM)
An engineered system cannot manage itself.
It must be managed.
System Reliability &Safety LaboratorySRSL 12
Health Management of Power Transformer
1 2 3 4 5 6
7 8 9 10 11 12
Health SensingHealth
ReasoningHealth
PrognosticsHealth
Management
1 3
2 4
Signal from Sensor 1
Signal from Sensor 2
Health Index for Failure Mode 6
Health Index for Failure Mode 9
Prognostics and Health Management (PHM)
PHM
Functions
System Reliability &Safety LaboratorySRSL 13
Intelligent Prognostics Platform for Wind Turbine Gearbox
Prognostics and Health Management (PHM)
Raw sensory data
Health Sensing &
Data Acquisition
Condition-Based Control
and Maintenance
11
11
Health-relevant
feature
Frequency spectrum
Data Processing &
Feature Extraction
22
22
Gear RULBearing RUL
500 1000 1500 2000 25000
1
2
3
4
5
6×10-3
RUL Distributions
Time (days)
Health
Prognostics
44
44
Wind turbine
gearbox
Health
Diagnostics
33
Feature 1
Fe
atu
re 2
Health state map33
Degradation mode 1
Healthy
Mode 2
System Reliability &Safety LaboratorySRSL 14
Research Timeline
Reliability-
Based Design
Fatigue Life
PD
F
Optimized
Initial
Limit
Prognostics and Health Management (PHM)
Sensing Reasoning Prognostics
Remaining Useful Life (RUL)
PD
F
Health Index 1
He
alt
h I
nd
ex
2
Battery Prognostics
Deep Brain StimulatorsLi-Ion Rechargeable
Scientist @ MDT / Faculty @ ISUPhD Student @ UMD
2007 2009 2011 2013 2015
System Reliability &Safety LaboratorySRSL 15
Li-Ion Battery in Implantable Medical Devices
Since 2004 Since 2010
Spinal Cord Stimulators
Mild electrical stimulation in the
spinal cord to alleviate chronic pain.
Deep Brain Stimulators
Targeted electrical stimulation to part of
brain for mitigating movement disorder.
Targeted longevity of 9 years, and 1000+ cycles
Inductively coupled recharge
System Reliability &Safety LaboratorySRSL 16
Need to know more:
� Capacity every recharge cycle
Battery Prognostics
Do Patients/Physicians Need to Know More?
Patients/physicians are informed
of battery charge level
Remaining
capacityBOL capacity80%
Remaining
useful life 6 8 years months
Neurostimulator
Patient programmer
with antenna
Therapy screen
� Remaining use life during
annual check-up
System Reliability &Safety LaboratorySRSL 17
Battery Prognostics
Schematic of Prognostics
Estimated capacity based on voltage and current measurements
Projected capacity Predicted end of life (EOL)
Particle filter used to
consider two sources of
uncertainty:
• Capacity estimation
• Model projection
70
80
90
100
0 100 200 300 400 500
No
min
al
Ca
pa
city
(%
)
Cycle Number
StatisticalLife Prediction
True Life
EOL limit
Hu C., Jain G., Tamirisa P., and Gorka T., “Method for Estimating Capacity and Predicting Remaining Useful Life of Lithium-Ion
Battery,” Applied Energy, v126, p182–189, 2014.
System Reliability &Safety LaboratorySRSL 18
Particle Filter for Estimating Joint Distribution of Model Parameters
[Pitt and Shephard, 1999, Journal of the American Statistical Association]
Step 1: Evaluate
Importance Weights
Step 2: Selection
Likelihood function based on capacity estimates
Step 3: Sampling
X = {k1, k2, t0, mc}
Q(t,C) = k1(1-e-t/to) + k2t + mcC
Battery Prognostics
Hu C., Jain G., Tamirisa P., and Gorka T., “Method for Estimating Capacity and Predicting Remaining Useful Life of Lithium-Ion
Battery,” Applied Energy, v126, p182–189, 2014.
System Reliability &Safety LaboratorySRSL 19
Battery Prognostics
0 [0] 200 [3.1] 400 [6.0] 600 [8.8]0
200
400
600
800
Cycle number [years on test]
RU
L (
cycle
s)
True RUL
Predicted RUL
0 [0] 200 [3.1] 400 [6.0] 600 [8.8]0
200
400
600
800
Cycle number [years on test]
RU
L (
cycle
s)
True RUL
Predicted RUL
Life Predictions @ Multiple Cycles
Hu C., Jain G., Tamirisa P., and Gorka T., “Method for Estimating Capacity and Predicting Remaining Useful Life of Lithium-Ion
Battery,” Applied Energy, v126, p182–189, 2014.
Life Prediction @ Cycle 200
Cycle number [years on test]
Norm
aliz
ed c
apacity (
%)
0 [0] 200 [3.1] 400 [6.0] 600 [8.8] 800 [11.5]70
75
80
85
90
95
100
Pre
dic
tio
n @
Cyc
le 2
00
Predicted
Life
Failure Limit
True Life
Real data
System Reliability &Safety LaboratorySRSL 21
Wind Energy
Prognostics
and Health
Management
Design for
Failure
Prevention
Future Research Plan
Reliability-
Based Design
Design for
Functional
ReliabilityDesign for
Resilience
Energy Storage Student to be identified
• Reliability evaluation and
failure prognostics of new
materials
Ms. Kayla Johnson
(PhD Student)
• Intelligent prognostics
of wind turbine
gearbox
System Reliability &Safety LaboratorySRSL 23
Data Processing Health Diagnostics Health Prognostics
Fast Fourier transform Self-organizing map Similarity-based interpolation
Wavelet analysis Clustering analysis Bayesian linear regression
Principle component analysis Mahalanobis distance Particle Filter / MCMC
Expert feature extraction Support vector machine Ensemble prognostics
Statistical correlation (copula) Relevance vector machine Semi-supervised learning
Artificial neural networks K-Nearest Neighbor
Classification Fusion
Extended Kalman Filter
PHM Toolbox being Developed at Hu’s Lab
System Reliability &Safety LaboratorySRSL 24
Journal Publications on PHM
1. Wang P., Youn B.D., and Hu C., “A Probabilistic Detectability-Based Sensor Network Design Method for System HealthMonitoring and Prognostics,” Journal of Intelligent Material Systems and Structures, DOI:10.1177/1045389X14541496, 2014. [ DOI ]
2. Hu C., Wang P., Youn B.D., and Lee W.R., “Copula-Based Statistical Health Grade System against Mechanical Faults ofPower Transformers,” IEEE Transactions on Power Delivery, v27, n4, p1809–1819, 2012. [ DOI ]
3. Youn B.D., Park K.M., Hu C., Yoon, J.T., and Bae Y.C., “Statistical Health Reasoning of Water-Cooled Power GeneratorStator Bars Against Moisture Absorption,” IEEE Transactions on Energy Conversion, vPP, p1–10, 2015. [ DOI ]
4. Hu C., Jain G., Schmidt C., Strief C., and Sullivan M., “Online Estimation of Lithium-Ion Battery Capacity Using SparseBayesian Learning,” Journal of Power Sources, v289, p105–113, 2015. [ DOI ]
5. Bai G., Wang P., and Hu C., “A Self-Cognizant Dynamic System Approach for Prognostics and Health Management,”Journal of Power Sources, v278, p163–174, 2015. [ DOI ]
6. Bai G., Wang, P., Hu C., and Pecht M., “A Generic Model-Free Approach for Lithium-ion Battery Health Management,”Applied Energy, v135, p247–260, 2014. [ DOI ]
7. Hu C., Jain G., Zhang P., Schmidt C., Gomadam P., and Gorka T., “Data-Driven Approach Based on Particle SwarmOptimization and K-Nearest Neighbor Regression for Estimating Capacity of Lithium-Ion Battery,” Applied Energy,v129, p49–55, 2014. [ DOI ]
8. Tamilselvan P., Wang P., and Hu C., “Health Diagnostics Using Multi-Attribute Classification Fusion,” EngineeringApplications of Artificial Intelligence, v32, p192–202, 2014. [ DOI ]
9. Hu C., Youn B.D., and Chung J., “A Multiscale Framework with Extended Kalman Filter for Lithium-Ion Battery SOC andCapacity Estimation,” Applied Energy, v92, p694–704, 2012. [ DOI ]
10. Hu C., Youn B.D., Kim T.J., and Wang P., “Semi-Supervised Learning with Co-Training for Data-Driven Prognostics,”Mechanical Systems and Signal Processing, v62–63, p75–90, 2015. [ DOI ]
11. Hu C., Jain G., Tamirisa P., and Gorka T., “Method for Estimating Capacity and Predicting Remaining Useful Life ofLithium-Ion Battery,” Applied Energy, v126, p182–189, 2014. [ DOI ]
12. Xi Z., Wang P., Rong Jing, and Hu C., “A Copula-Based Sampling Method for Data-Driven Prognostics,” ReliabilityEngineering and System Safety, DOI: 10.1016/j.ress.2014.06.014, 2014. [ DOI ]
13. Hu C., Youn B.D., and Wang P., “Ensemble of Data-Driven Prognostic Algorithms for Robust Prediction of RemainingUseful Life,” Reliability Engineering and System Safety, v103, p120–135, 2012. [ DOI ]
14. Wang P., Youn B.D., and Hu C., “A Generic Probabilistic Framework for Structural Health Prognostic and UncertaintyManagement,” Mechanical Systems and Signal Processing, v28, p622–637, 2012. [ DOI ]
Health
Sensing &
Data
Processing
Health
Diagnostics
Health
Prognostics