prognostics implementation in aerospace applications · emerson appliance solutions ... rendell...
Post on 08-Apr-2018
220 Views
Preview:
TRANSCRIPT
University of MarylandCopyright © 2009 CALCE
1calceTM 1Prognostics and Health Management Group
Prognostics Implementation in Aerospace Applications
PrognosticsTM
Michael Pecht, Ph.D.CALCE Electronic Products and Systems
University of Maryland – USA
Prognostics and Health Management, Condition-Based Maintenance and Health
& Usage Monitoring Symposium21-22 April, 2009
University of MarylandCopyright © 2009 CALCE
2calceTM 2Prognostics and Health Management Group
• Formally started in 1984, with support from NSF, as a US Center of Excellence in electronics systems reliability.
• Over $6.5M in funding per year, by over 150 of the world’s leading electronics organizations
• One of the world’s most advanced and comprehensive electronics testing and failure analysis laboratories
• Supported by 112 faculty, visiting scientists and research assistants
A Brief History of CALCE
University of MarylandCopyright © 2008 CALCE
Prognostics and Health Management 3
• Alcatel-Lucent• ALZA (Johnson&Johnson)• Amkor• Arbitron• Arcelik• ASC Capacitors• ASE• Arbitron• Astronautics• Atlantic Inertial Systems• AVI-Inc• Axsys Engineering• Battelle • Branson Ultrasonics• Brooks Instruments• Capricorn Pharma• Cascade Engineering • AMSAA Reliability Branch• Boeing• BAE• CAPE – China• CDI• Cisco Systems, Inc.• Crane Aerospace & Electronics• Curtis Wright/Dy4 De Brauw
Blackstone Westbroek• Defense Micro Electronics
Activity• Dell Computer Corp.• EIT, Inc.• Embedded Computing &
Power• EMCORE Corporation
• EADS – AirBus• Emerson Advanced Design• Emerson Appliance Controls• Emerson Appliance Solutions• Emerson Electric Co.• Emerson Network Power• Emerson Process Management• Ericcson• DRS EW Network Systems,
Inc.• Essex Corporation • Exponent, Inc.• Fairchild Controls Corp.• Filtronic Comtek• GE Fanuc Embedded Systems• GE Global Research• General Dynamics – AIS• General Motors• Guideline• Hamlin Electronics Europe• Hamilton Sundstran• Harris Corp• Honda• Honeywell• Howrey, LLP• Huawei• Intel• Juniper• Kimball Electronics• L-3 Communication Systems
• LaBarge, Inc• Laird Technologies • Liebert Power and Cooling• LM Aero - Ft. Worth Site• Lockheed Martin Lutron
Electronics • Maxion Technologies, Inc.• Motorola• Joint Strike Fighter
Program• Mobile Digital Systems, Inc.• n-Code• NASA Goddard Space
Flight• NetApp• Nokia• Northrop Grumman• NXP Semiconductors• Ortho-Clinical Diagnostics• PEO Integrated Warfare• Petra Solar • Philips• Philips Medical Systems• Pole Zero Corporation• Pressure Biosciences• Raytheon • Rendell Sales Company• Research in Motion• RNT, Inc.• Rolls Royce• Rockwell Automation
• Samsung Memory• Samsung Techwin• S.C. Johnson Wax• SanDisk• Schlumberger• Schweitzer Engineering Labs • Sensors for Medicine and
Science, Inc.• SiliconExpert• SolarEdge Technologies• Space Systems Loral• Starkey Laboratories, Inc• Sun Microsystems• Symbol Technologies, Inc• Team Pacific Corporation• Tech Film• Tekelec• Teradyne• The Bergquist Company• The M&T Company• The University of Michigan• Tin Technology Inc• TruePosition, Inc.• TÜBİTAK Space Technologies• Vectron International, LLC• Weil, Gotshal & Manges LLP• Whirlpool Corporation• WiSpry, Inc.• Woodward Governor
Organizations that Fund CALCE: $6.5M in 2008
University of MarylandCopyright © 2009 CALCE
4calceTM 4Prognostics and Health Management Group
RAMS Challenges in Aerospace
• Key systems are now controlled by a complex network of electronics– Guidance, Navigation, and Control– Communications and Tracking– Sensors and Instrumentation– Electrical Power Systems– Data Processors– Data Busses/Networks
University of MarylandCopyright © 2009 CALCE
5calceTM 5Prognostics and Health Management Group
Challenges for Avionics Industry
• COTS components• Rapidly changing technology / faster obsolescence • Imperfect screening and qualification standards• Intermittent failures• Significant numbers field failures turn out to be NTF• Redundant systems• Scheduled maintenance
University of MarylandCopyright © 2008 CALCE
6calceTM 6Prognostics and Health Management
Environmental Demands on Avionic Systems• high temperatures: 110 °C for 2000 h
• low temperatures: -60 °C for 100 h / -55°C 200 h
• temperature cycle: -40 °C / 110 °C 2000 h
• temperature cycle (slow):
-40/+85 °C, soak time Tmin 90 min, Tmax 150 min, tchange 150 min
• temperature shock: -40/+85 °C, soak time Tmax 30 min, Tmin 45 min, tchange 10 s
• temperature storage: 85 °C, 680 h
• relative humidity (cyclic):
55 °C 93% r.h, 25 °C 95 % r.h., cycle duration 4 h, 6 days
• low pressure test: 91,2 mBar 100 h
• radiation Xenon-light:
3000 h
• radiation UV-light: 3000 h
University of MarylandCopyright © 2008 CALCE
7calceTM 7Prognostics and Health Management
Worst CaseUse Conditions Typical Use Conditions
Use categoryMin temp (°C)
Max temp (°C)
Service life
(years)
Tmin(°C)
Tmax(°C)
ΔT (°C) Hours Yearly
cycles
Consumer 0 60 1-3 20 55 35 12 365Computer 15 60 5 25 45 20 2 1460Telecom -40 85 7-20 10 45 35 12 365Commercial aircraft -55 95 20 20 40 20 12 365Industrial/ auto -55 95 10 30 50 20 12 185Military ground/ship -55 95 10 5 45 40 12 100Space -55 95 5-30 20 55 35 1 8760Military avionics -55 95 10 0 80 80 2 365Auto (under hood) -55 125 5 20 80 60 1 1000IPC-SM-785, Guidelines for Accelerated Reliability Testing of Surface Mount Solder Attachments, November 1992.
Beware of “Standard” Profiles
University of MarylandCopyright © 2008 CALCE
8calceTM 8Prognostics and Health Management
Instrument Panel
EE Bay Air
EE Bay (AEM)*Console (AEM)*
Dallas San Diego
Ren
oPhoenix
Vega
sR
eno
Vega
s
SMF
Phx Milwaukee Phx
60
50
40
30
20
10
0
-10
0.00
3.59
5.58
7.57
9.56
11.5
513
.54
2.00
15.5
3
19.5
121
.50
23.4
91.
483.
475.
46
17.5
2
7.45
11.4
313
.42
15.4
117
.40
19.3
921
.38
9.44
Time (hh.mm)
Tem
pera
ture
(°C
)AvionicsFan
Jet Engines
Example: Aircraft Temperature Profile
Cluff, K., Barker, D., Robbins, D., and Edwards, T., “Characterizing the Commercial Avionics Thermal Environment for Field Reliability Assessment,” Journal of the IES, Vol. 40, No. 4, pp. 22-28, 1997.
* Measured by an Aircraft Equipment Monitor (AEM) fitted in a commercial aircraft.
University of MarylandCopyright © 2008 CALCE
9calceTM 9Prognostics and Health Management
EADS HUMS and CALCE Prognostics ProjectMicrocontrollerTemperature & RH sensor Terminals for data
communication
FRAM memoryTerminals for external sensors
Terminals for battery
2-axis accelerometer
RFID
Courtesy of EADS
University of MarylandCopyright © 2008 CALCE
10calceTM 10Prognostics and Health Management
CALCE E-Prognostics Sensor Tag
University of MarylandCopyright © 2008 CALCE
11calceTM 11Prognostics and Health Management
Inspection Intervals• IL-Check - 48 month
– Detailed inspection of structure, fuselage and wings. – Check and maintenance of electronic and hydraulic systems – Implementation of improvements, complete cabin
maintenance• D-Check - 72 month
– General maintenance, check of panels, bolts, screws for fatigue or wear out.
– Replacement of bigger parts, – Replacements of all instruments and devices, – Duration: 4-6 week, extend: 30,000- 50,000 working hours
University of MarylandCopyright © 2008 CALCE
12calceTM 12Prognostics and Health Management
PHM Challenges in Avionics
• The cost of maintenance is already over 15% of the total operation costs
• System / component faults and failures are very difficult to detect, diagnose, and mitigate in-flight with existing technologies.
• Need Prognostics and Health Management (PHM)
University of MarylandCopyright © 2008 CALCE
13calceTM 13Prognostics and Health Management
Prognostics-Based Logistics Demo: F/A-18
LRU with PHM Sensor Module and RFID
Mission data and id
Net-Centric logistics database
Smart Reader/Writer
Logistics Planner
LRU history and prognostics
Queries and updates for logistics decisions
Updated logistics / maintenance history and id
In 2005 UMD team was awarded $2.1M from US OSD to develop an interactive supply chain that is based on prognostics, RFID, and network databases
University of MarylandCopyright © 2009 CALCE
14calceTM 14Prognostics and Health Management Group
NASA Contract: Reliable Diagnostics and Prognostics for Critical Avionic Systems
• Project objectives include:– Develop approaches to detect faults, to model degradations,
and to predict failures in avionics components. – Develop a methodology involving parameter selection, feature
extraction, pattern recognition, anomaly detection, parameter isolation, and remaining useful life estimation.
– Equip NASA with the ability to monitor the health of onboard electronics of an aircraft in actual operating conditions to increase the safety and availability of the aircraft.
University of MarylandCopyright © 2009 CALCE
15calceTM 15Prognostics and Health Management Group
CALCE PHM Methodology
• Existing sensor data• Bus monitor data• BIT, IETM
SystemPrognosticsAnd Health Monitoring
Remaining Life Assessment
CALCE – ePrognosticsSensor System
Life Cycle Logistics and Cost Analysis
Virtual Life Estimation
Physics-of-Failure Based Approach
Data Driven Approach
Failure Modes, Mechanisms and Effects Analysis (FMMEA)
Maintenance Records
Failure Mechanisms
Failure Modes
Life Cycle Profile
Physics of Failure Models
Design Data
Detection, Severity & Occurrence
Fusion Prognostics
University of MarylandCopyright © 2009 CALCE
16calceTM 16Prognostics and Health Management Group
CALCE Fusion Prognostics (simplified)Healthy Baseline
Continuous Monitoring
Parameter Isolation
Alarm
Remaining Useful LifeEstimation
PoF Models
Anomaly?
Identify parameters
Data Driven Algorithms
FailureDefinition
Yes
No
Database and Standards
University of MarylandCopyright © 2009 CALCE
17calceTM 17Prognostics and Health Management Group
Package Interconnections
EMIsusceptibility
EMIgeneration Crosstalk
Circuitry
Excessivedelay time DC drop
Connections and Connectors
DI noise
Connector corrosion
PTH barrelfatigue
Lead padcorrosion
Tracecorrosion
Tracefracture
Laminateplasticization
Delamination
Tglimitation
Fiber resindebonding
CFF
Dendriticgrowth
Intermetallicformation
GullwingLow cycle fatigueHigh cycle fatigue
Shock fracture
BGALow cycle fatigueHigh cycle fatigue
InsertionPullout
Lead fatigueHigh cycle fatigue
Shock fracture
COBLow cycle fatigueHigh cycle fatigue
LCCLow cycle fatigueHigh cycle fatigue
Shock fracture
J-leadLow cycle
fatigueHigh cycle
fatigueShock fracture
Pressure contactSpring relaxation
Pin in socketPin fretting
Edge cardFinger fretting
Printed Wiring Board
Failure Sites and Mechanisms in Electronic Hardware
University of MarylandCopyright © 2009 CALCE
18calceTM 18Prognostics and Health Management Group
CALCE Probabilistic PoF Prognostics
,.....),,,( Dmean tdtdssfw s∆=∆Damage,
Time (t)
Load
(s
)
Embedded Data Reduction and Load Parameter Extraction
Remaining life = 1 - g(Σ∆w)
Mean load (Smean) Ramp rate (ds/dt)0
0.25
0.5
Range (∆s)
Freq
uenc
y
Dwell time (tD)0
0.25
0.5
0
0.25
0.5
0
0.25
0.5
0
10
20
30
40
50
60
70
80
90
100
7/19/0612:00 AM
7/20/0612:00 AM
7/21/0612:00 AM
7/22/0612:00 AM
7/23/0612:00 AM
7/24/0612:00 AM
7/25/0612:00 AM
7/26/0612:00 AM
7/27/0612:00 AM
7/28/0612:00 AM
Deg
rees
C
University of MarylandCopyright © 2009 CALCE
19calceTM 19Prognostics and Health Management Group
Physics of Failure Methods Work Well HerePhysics of Failure Methods Work Well HereData-Driven Methods Work Well HereThe monitored data may not directly relate to a specific
failure mechanism or to inputs to a failure model
University of MarylandCopyright © 2009 CALCE
20calceTM 20Prognostics and Health Management Group
Data Driven Analysis Procedure
Acquire new observations
Create “healthy”
profile matrix Calculate expectations
Calculate actual residuals
RX=Xexp-XobsSelect
parameters to
monitorAssess
variability wrt other healthy
data
Calculate expectations
Calculate healthy residuals
RL =Lexp-L
TrendAnalysis
University of MarylandCopyright © 2009 CALCE
21calceTM 21Prognostics and Health Management Group
CALCE Fusion Prognostics (simplified)Healthy Baseline
Continuous Monitoring
Parameter Isolation
Alarm
Remaining Useful LifeEstimation
PoF Models
Anomaly?
Identify parameters
Data Driven Algorithms
FailureDefinition
Yes
No
Database and Standards
University of MarylandCopyright © 2009 CALCE
22calceTM 22Prognostics and Health Management Group
Case Study #1: Circuit Card AssemblySubjected to Temperature Cycling
University of MarylandCopyright © 2009 CALCE
23calceTM 23Prognostics and Health Management Group
PoF Approach for Time to Failure Estimation
• A failure modes mechanisms and effects analysis (FMMEA) was carried out to determine the critical modes and mechanisms that affect the components
• Models for each identified critical failure mechanism were selected to calculate the time to failure from a database of PoF models.
• The mean cycles to failure for the first to fail component: 256 I/O BGA was calculated to be 1038 cycles
• The 3 sigma cycles to failure: 256 I/O BGA was calculated to be 720 cycles
University of MarylandCopyright © 2009 CALCE
24calceTM 24Prognostics and Health Management Group
Anomaly Detection
• A healthy baseline was established for each type of component using the first 100 thermal cycles of the test
• Estimates of the continuously monitored data were - assessed with a multi-variate state estimation
algorithm, - the residuals were statistically tested for
anomalies using SPRT, - the parameters causing an anomaly were
identified for assessment from the PoF database and the definition of failure from the PoF model
University of MarylandCopyright © 2009 CALCE
25calceTM 25Prognostics and Health Management Group
SPRT Signaled Alarms Starting at the 583rd
cycle
University of MarylandCopyright © 2009 CALCE
26calceTM 26Prognostics and Health Management Group
• The physics-of failure database analysis (from the data-driven analysis) provided a definition of failure for the resistance: 300Ω.
• Peak residual values from the time of anomaly detection was trended using regression analysis and the time taken for the resistance residuals to cross the 300Ω threshold was calculated.
• The estimates were updated every cycle• The time to failure, calculated based on the parameter trending
(resistance) was estimated to be 620 cycles. • This estimate was made at the 601st cycle (component failed at
the 631st cycle).
Data Driven Prognostics for RUL Estimation
University of MarylandCopyright © 2009 CALCE
27calceTM 27Prognostics and Health Management Group
RUL Estimates vs Product Failure
• Deterministic PoF mean cycles to failure estimate (t = 0): 1038 cycles (mean)
720 cycles (3 sigma)• Anomaly detected (possible intermittent failures)
583 cycles• The Fusion Prognostics cycles to failure estimate
(t = 601): 620 cycles• Actual failure 631 cycles
University of MarylandCopyright © 2009 CALCE
28calceTM 28Prognostics and Health Management Group
Comment on Prognostics-Based Qualification
and Screening
University of MarylandCopyright © 2009 CALCE
29calceTM 29Prognostics and Health Management Group
SPRT Signaled Alarms Starting at the 583rd
cycle
University of MarylandCopyright © 2007 CALCE
Center for Advanced Life Cycle Engineering 3030
• Fusion Prognostics look promising for prognostics
• PHM will be used for qualification, screening and continuous remaining life assessment.
• Prognostics will be embedded in most critical electronics within the next 10 years
Conclusions
top related