An Unified Approach to Structural Health Management and Damage Prognosis of Metallic Aerospace...
Transcript of An Unified Approach to Structural Health Management and Damage Prognosis of Metallic Aerospace...
An Unified Approach to Structural Health Management and Damage Prognosis of Metallic
Aerospace Structures
Prognosis of Aircraft and Space Devices, Components, and SystemsAir Force Office of Scientific Research
University of Cincinnati, Cincinnati, Ohio
February 19 and 20, 2008
Grant Number: FA95550-06-1-0309
Program Manager: Dr. Victor Giurgiutiu
Aditi Chattopadhyay
Department of Mechanical and Aerospace Engineering
Arizona State University
2
MURI
Structural Health Monitoring & Prognosis of Aerospace Systems
Douglas Cochran•Statistical signal processing•Theory of sensing•Mathematical modeling
James B. Spicer•Materials process monitoring & control
•Ultrasonics
•High-temperature characterization
Dan Inman•SHM
• Wireless sensing and damage assessment
•Membrane optics
Roger G. Ghanem•Risk assessment•Stochastic mechanics•Computational mechanics•Inverse problems and optimization
Antonia Papandreou•Signal processing •Sensing & Information Processing •Detection & Estimation
Aditi Chattopadhyay•Smart structures
•SHM
•Multiscale modeling
•Mechanics of composites
•MDO
Pedro Peralta
•Fracture & fatigue •Composite materials •FEM •Continuum mechanics
MURI RESEARCH TEAMASU TeamAcademic Professionals: Jun Wei, Narayan Kovvali
Graduate Students : Debejyo Chakraborty, Clyde Coelho, Chuntao Luo, Subhasish Mohanty, Donna Simon, Sunilkumar Soni, Rikki Teale, Christina Willhauck
USC TeamAcademic Professional: Maarten Arnst
Graduate Students : Maud Comboul, Sonjoy Das, Arash Noshadravan
JHU TeamAcademic Professional: Seyi Balogun
Graduate Students : Lindsay Channels, Travis DeJournett
VT TeamAcademic Professional: Benjamin L. Grisso
Graduate Student : Mana Afshari
BENEFIT TO DOD/INDUSTRY TECHNICAL APPROACH
•Physically based models to characterize damage nucleation & growth
•Characterize wave propagation in hotspot
•Optimally integrate sensor network
•Waveform design & damage detection
•Sensor management schemes for detection/classification
•Stochastic models to account for uncertainties
•Estimation of remaining useful life
•Validation on test structures
• Connect microscopic damage to macroscopic scale monitoring
• Sensor sensitivity
• Sensor/host structure coupling
• Hierarchical information management
• Hybrid approach for life estimation
• Precursor to damage/first failure to inspection
OBJECTIVES•Computationally efficient multiscale modeling techniques for characterizing the damage state of a material (including nucleation and growth)
• Damage detection and classification techniques for sensor integration and instrumentation
• Prognosis capabilities for predicting failure probability and remaining useful life
• Testing, validation and application
BASIC RESEARCH ISSUES
• Improved techniques will facilitate assessment of health of metallic aircraft structures
• Project outcome will help surmount some of the technical challenges, complementing ongoing activities at AFRL
• Research results will help establish improved IVHM systems
• Future aircraft systems can benefit from integration with prognosis programs focused on current aircraft.
• Advancements in damage analysis, detection & classification are useful sustainable infrastructure and electronic system monitoring
TASK DESCRIPTIONS AND PERSONNEL
A. ChattopadhyayP. Peralta
Roger Ghanem
Task 1•Material Charecterization•Multiscale model to predict damage nucleation & growth
Task 2•Optimal sensor placement•Detection•Signal processing•Diagnosis & classification
Task 3•State awareness •Life prediction
Task 4 (All PI’s)Testing, validation and
applications
A. ChattopadhyayP. Peralta
James Spicer
A. ChattopadhyayA. Papandreou-Suppappola
Daniel InmanDouglas Cochran
AFRL / VA Boeing Phantom Works
DOD COLLABORATIONS AND TRANSITION TO REAL SYSTEMS
• Collaboration with AFRL: • Mark Derriso, Structural Health Assessment Team Leader, AFRL/VA
• Provides data from AFRL experimental set-ups• Frequent meetings with Mark and his team: discuss MURI progress
and relevant AFRL problems needed to help transition of our work to real systems
• Meetings with Jim Larsen (AFRL/MLLMN) and Kumar Jata (AFRL/MLLP) • Collaboration with Boeing Phantom Works (Eric Haugse)
• Hotspot Program with AFRL (involves actual F-18 testing in Arizona for transition to real systems).Participants: AFRL, Boeing Phantom Works, Accelent Technologies, Metis Design
• HotSpot Monitoring Program teleconference (bi-weekly)• Advisory board committee provides feedback:
• Members from AFRL, US Air Force Academy, United Technologies Research Center, Boeing, Next Generation Aeronautics, Lockheed Martin Aeronautics Company, National Transportation Safety Board, NRL, Naval Surface Warfare Center, NASA GRC, NASA LaRC, NASA ARC, US Army ARDECOM, Los Alamos National Lab.
RVE for Grains/ Particles3D Grain/ Particle size
distributions
Material
Characterization
Multiscale
ModelingMicrostructure Reconstruction
Short Crack Growth in the Mesoscale
Structure Level Fatigue Simulation
Representative Microstructure (FEM)
Metallography Microscale
Damage Initiation
Physically-based Multiscale Modeling
TECHNICAL APPROACH
Strain fields ahead of fatigue cracks in wrought Al alloys: in-situ testing and DIC
Nanoindentation of precipitates in wrought Al alloys
Load stage
Loaded specimen
Load Direction and Rolling Direction (RD)
MATERIALS CHARACTERIZATIONMultiscale Material Characterization
Crack tipCrack tip
Use Electron Backscatter Diffraction (EBSD) along with serial sectioning: 2-, 2.5- and 3-D
“Artificial” microstructures are also being generated
Same grain size (100 µm) different grain size distribution Large (300 µm) grain size
2.5-D 3-D
MATERIALS CHARACTERIZATIONMicrostructure Reconstruction and Representation
2-D
Use microstructure representation and meshing tools: defects can be included
Results show effects of microstructural variability on local fields
2.5-D 3-D
MATERIALS CHARACTERIZATIONMicrostructurally Explicit Finite Element Models
2-D
2-D
3-D
11
INTERACTION OF RELEVANT SCALES IN MULTISCALE MODELING
2-D Slice
2.5-D Representation
3-D Representation
Material Characterization
Void Model
Single Crystal Structure
Micro Scale
Crystal Orientation
Crystal Properties
Orientation Distribution &
Properties
Short Crack Propagation
Meso Scale
Crack Initiation
Polycrystal Structure
Homogenization
Localization
Wave Propagation
Long Crack Propagation
Macro Scale
Component
Damage Parameter
0. 00
0. 01
0. 01
0. 02
0. 02
0. 03
0. 03
0. 00 5. 00 10. 00 15. 00 20. 00 25. 00 30. 00 35. 00
Time (s)
Dam
age
Par
amet
er
0
50
100
150
200
250
300
0 0.002 0.004 0.006 0.008 0.01
Strain
Str
ess
(MP
a)
Hardening Parameter
12
-400
-300
-200
-100
0
100
200
300
400
-0.004 -0.003 -0.002 -0.001 0 0.001 0.002 0.003 0.004
0.0E+00
2.0E-02
4.0E-02
6.0E-02
8.0E-02
0 2 4 6 8
Mises stress distributionStress-strain response
FATIGUE ANALYSIS (SINGLE CRYSTAL)
Str
ess
(MP
a)
Strain
Acc
um
ula
ted
sh
ear
stra
in
Number of cycles
Capture crystal orientation
Fatigue hardening & saturation
Accumulative shear strain
Mesoscale
13
-4 -3 -2 -1 0 1 2 3 4
Strain X103
-500
-200
-300
-400
-100
100
0
200
300
400
500
Str
es
s (
MP
a)
Grains
OIM (Orientation Imaging Microscopy)
Scan
MESOSCALE STRUCTURE
OOF
ABAQUS & UMAT
Methods for In Situ Interrogation and Detection
Sensor design, network and placement
Damage detection and classification
Nonlinear ultrasonicdamage characterization
Sensing multi-scale damage with impedance, vibration,
& Lamb wave based methods
Time-frequency & statisticaldamage classification: AFRL TPS, ASU bolted-joint data
Mesoscopic ultrasonic techniques for assessment of material microstructure
FEM based analysis of macro- length scale damage
with virtual sensors
Bayesian sensor fusion of data received from
multiple distributed sensors
TECHNICAL APPROACH
-0.00636
-0.00477
-0.00318
-0.00159
0
-0.05
-0.04
-0.03
-0.02
-0.01
0
0 400 800 1200 1600
Am
plit
ude o
f M
od
el
Am
plitu
de o
f Un
fatig
ued 6
06
1
Time (ns)
Experimental Schematic for Laser Ultrasonic Investigations
Nd:YAG 1064 nm 9 ns pulse
Iris
Sample on translation stage
Nd:YAG532 nmcontinuous
Stabilization circuit
Lens
Piezoelectric mirror mount
Oscilloscope
Ultrasonic generation
2.4 mJ
Receiver
Lens
Mirror
Michelson type interferometer
+
_
Surface displacement
ASU SP Group (Papandreou, Cochran)
JHU Ultrasonics Group (Spicer)
data
Ultrasonic displacement measured at the epicenter
Model
Measurement
Spectrogram
Team Integration
RESULTS: NONLINEAR ULTRASONICS
ResultsSample Tested
Lug joint: typical structural hot spot
Actuator
Sensor
ASU SP Group (Papandreou, Cochran)
ASU Modeling Group (Chattopadhyay)
features for SVM
Team Integration
RESULTS: SUPPORT VECTOR MACHINE BASED DAMAGE CLASSIFICATION
Collaboration with Mark Derriso (AFRL/VA)
• Damage Class definition: - Class 1 = Bolt 1 at 25%
torque- Class 2 = Bolt 2 at 25%
torque - Class 3 = Bolt 3 at 25%
torque- Class 4 = Bolt 4 at 25%
torque- Class 5 = All bolts at 100%
torque (fully tightened/healthy case)
• PZTs attached to bolted square aluminum plate
• PZT-1 used for transmitting 0-1.5 kHz chirp
• Signals received at PZT-2, PZT-3, and PZT-4
S. Olson, M. DeSimio, and M. Derriso, “Fastener Damage Estimation in a Square Aluminum Plate”, Structural Health Monitoring Journal, 2005
Confusion matrix (HMM based damage classifier)
0.80500 0.05500 0.13500 0 0.00500
0 0.98000 0.00500 0 0.01500
0.05500 0.08000 0.86500 0 0
0.00500 0.02000 0 0.97000 0.00500
0 0 0 0.00125 0.99875
RESULTS: TIME-FREQUENCY CLASSIFICATION
Probabilistic Data Driven Prognosis Model
Fracture Mechanics Based Physics Model
System Identification Prognosis Model
Prediction of crack growth and plastic zone parameters by
Gaussian Process Model
Prediction of effective stress intensity factors that account
for closure effects
Vibration and wave based system identification for damage state estimation
R
U
L
E
Hybrid
Prognosis Model
Prognosis via State-Awareness and Life Models
TECHNICAL APPROACH
Flight Cycle
Dam
age
Ind
ex (
Cra
ck L
eng
th)
k N N+1
GAUSSIAN PROCESS DATA DRIVEN APPROACH• Based on high dimensional kernel function
• Uncertainty quantified using Bayesian approach
• History as training distribution
• Predicts new mean damage and associated variance
• Predicts possible collapse point if new predicted variance
exceeds threshold flag
DATA DRIVEN MODEL PREDICTION
Single Variate Model
Prediction
Load Spectrum
HYBRID PROGNOSIS APPROACH
Data driven model for calculating
plastic zone constraint factor
Incremental crack length from
physics based model
Results From Pure Physics Based Model Results From Hybrid Model
•Detection•Signal processing•Diagnosis & classification
Material characterization, multiscale model and state awareness model
Testing, Validation & Application
Calibrate and validate modeling methods
Sensor network and placement
Application to Structural Hotspots
AFRL/VA, Boeing Structural Hotspot
Program
TECHNICAL APPROACH
Sample 1 (Polished) Sample 2 (Sand Blasted) Sample 3 (Sand Blasted)
Life of sample 2 about twice of sample 1 under similar loading condition
Two distinct damage nucleation sites for sample 2
Failure mode - High cycle fatigue from shoulders for sample 1 & 2
- Very high cycle fretting fatigue from pin hole
Induced Stresses Influence Fatigue Life and Failure Patterns
EXPERIMENTAL OBSERVATIONS
380,621 823,537 >3 MillionCycles to failure (110 – 1100 lbs) (80 – 800 lbs)(110 – 1100 lbs)
GUIDED WAVE IN LUG JOINT
Healthy
Damaged
FUTURE WORKTask 1:• Predict damage nucleation & propagation using modified fatigue damage criteria.• Simulate sensor signals & study their interaction with cracks using distributed point
source method (DPSM) – a wave based approach.
Task 2:• Adaptive signal processing and classification using active and multi-task learning
methodologies.• Use of data from new sensors and physics based FEM modeling to train damage
detection and classification algorithms.
Task 3:• Formulate multivariate prognosis models that incorporate physical-based models to
account for load sequence effects.• Incorporate material and sensor signal variability into prognosis framework.• Develop a prognosis approach for crack nucleation based on "virtual sensors" (output
from multiscale modeling) to estimate life spend to grow "detectable" damage.
Task 4:• Perform testing on instrumented samples with complex geometry (lug joints, bolted
joints) to gather statistical information on failure modes, sensor performance and to collect data for model validation (integration with Tasks 1, 2, and 3).
• Develop a test article for use with the biaxial load frame to obtain statistical information under both complex geometries and complex loading (integration with Tasks 1, 2, and 3).