development of advanced risk assessment...
TRANSCRIPT
Development of Advanced Risk Assessment Methodologies for Aircraft Structures Containing MSD/MED
M. Liao, Y. Bombardier, G. Renaud, N. Bellinger, T. Cheung (DTAES/DND)
Structures and Materials Performance LaboratoryInstitute for Aerospace Research
2
Acknowledgements
This work was performed with financial support from the DRDC-NRC collaborative project “Quantitative Risk Assessment of CF Aircraft Structures”
Project members:
Dr. G. Renaud, Mr. Y. Bombardier, Dr. M. Khan, Dr. G. Li, Dr. M. Liao
Dr. A. Fahr, Mr. N. Bellinger
DND support:Mr. K. McRae of DRDC
Mr. T. Cheung, Mr. Y. Caron, Mr. J. Gaerke of DTAES
Capt. T.J. Cadeau, Sgt. M. Bunn of ATESS/DND
3
Contents
• Risk Management for CF Air Fleets• NRC Risk Analysis Methods/Tools• MSD Damage Tolerance Analysis
– MSD/MED crack growth analysis– MSD/MED residual strength analysis
• Risk Analysis for MSD/MED Structures– ICSD/EIFSD– Monte Carlo MSD crack growth analyses– Maximum Stress Distribution
• Probability of Failure (PoF) Results• Concluding Remarks• Future work
4
Risk Management for CF Air Fleets
RARM (Record of Airworthiness Risk Management)• Hazard Id.� Risk Ass.� Risk Ctrl. �RARM Approval� Risk Tracking• Affecting all CF fleets (DND-AD-2007-01)
When “sufficient” data is available, Quantitative risk assessment (QRA) substantiates the assignment of a risk number in Qualitative risk assessment
TAM, C-05-005-001/AG-001,
DTAES/DND, 2001
TAM, C-05-005-001/AG-001,
DTAES/DND, 2001
TAM, C-05-005-001/AG-001,
DTAES/DND, 2001
5
• NRC developed methods and tools to calculate the single flight hour probability of failure (PoF, ~hazard rate) based on extensive durability and damage tolerance analysis (DaDTA)and stress-strength interference model
NRC Risk Analysis Methods
hourflight per on distributi stress maximum theis ][ where
)]([1)( :criterionstrength residualFor
)]),([1)(()(:criterion For
)()()]or ,([)(
0
0
σ
σ
σ
σσσ
σ
σ
σ
H
aHaPOF
dKKaHKfaPOFKc
daaPOFafKaPtPoF
RS
CCCCK
RSCCriticalMax
C
−=
−=
=≥=
�
�∞
∞
• Crack size distribution update based on NDI and repair
),( )](1[ ),(),()(),( ,
0 ,, tafaPODtafdatafaPODtaf beforeaRCSDbeforeaaftera −+⋅= �∞
6
NRC Risk Analysis Tools ProDTA
• ProDTA calculates the PoF using probability integration method or Monte Carlo technique
• ProDTA is under development, aiming to become a tool for CF fleets
Maximum stress(Gumbel / others)
Initial crack size distribution
(ICSD/EIFS)
Crack growth curve and β-solution
NDI POD(Log-logistic / others)
Failure criteria(KC, ac, σRS)
ProDTA
Maximum pit depth(Gumbel)
Corrosion growth rate(Weibull / database)
Corrosion protection breakdown time
(Normal)
Corrosion POD/NDI error
(Normal)
PoF
Fatigue inputs Corrosion inputs
Re. ICAF 2005 paper
7
Case Study: CC-130 Centre Wing MSD/MED Issue
The crisis
C-130A catastrophic failure in Walker, CA. 2002
The causes
“fatigue cracks in the lower wing skin” and “multiple site fatigue damage/ MSD” (NTSB)
The method neededAdvanced DaDTA and Risk Assessment Methodologies for Aircraft Structures Containing MSD/MED
8
CC-130 Center Wing Lower Surface Panel
CC-130 Center Wing, Lower Surface Panel, Location CFCW-1
Standard Crack (SC) scenario: single dominant crack, phase-by-
phase (PBP) approach (OEM DTA)
∅ 0.339” (BBR=1.587)
∅ 0.267”
7075-T7351 0.22” thick
VIII VII VI V IV III I II
Phases I & II
Phases III & IV
Phases V & VI
Phases VII & VIII
Multi-phase single crack growth analysis:
SC-PBP (OEM analysis, duplication)
9
Crack Growth AnalysisScenarios
MSD scenario: MSD approach
Standard Crack (SC) scenario:MSD approach
Primary crack(0.050”)
Secondary cracks(0.005”)
SC-MSD MSD
Primary crack(0.050”)
Secondary cracks(0.005”)
10Good agreement between NRC closed-form
equations, OEM, and FEA (StressCheck)
ββββ-Library
• Currently available & validated �-functions:
c
a
φ
Thickness (T)
B
σbearing
σtotal
W
c
σbypass
BBR=σbearing/σbypass
D B
σtotal
W
c
σtotal
B
D=2R2c
Corner crack
Radially crack at hole with bearing load
σtotal
Load path
Plate
Crack
Stiffener
c
Ligament failure
Stringer/Cap effect
Edge crack through hole
Crack approaching a
hole
11
ββββ-Library
• Additional available & validated �-functions:
2a1 2a2
A B C D
b
Gap
B
σbearing
σtotal
W
c1
σbypass
c2
BBR=σbearing/σbypass
B1
σtotal
W
c1 c2
σtotal
B2
D2D1
W
ci c2
σtotal * W/(W-Σci)
σtotal * W/(W-Σci)
Diametrically cracks at hole with bearing load
Crack interactioneffect
Linked-up crack Net section effect(under investigation)
Good agreement between NRC closed-form equations, OEM, and FEA (StressCheck)
12
Verification of MSDββββ-Solutions
MSD ����-solution from a benchmark MSD problem was verified with FEA (StressCheck) results (ICF12 paper, Ottawa, 2009)
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
2
2.1
2.2
0 10 20 30 40 50
a0 (mm)
ββ ββ-so
lutio
n
CGCC130MSD (A11)CGCC130MSD (A12)STRESSCHECK (A11)STRESSCHECK (A12)
a12 and a21 merged
a22 and a31 merged
1
1.1
1.2
1.3
1.4
1.5
1.6
1.7
1.8
1.9
2
2.1
2.2
50 100 150 200 250 300
a0 (mm)
ββ ββ-so
lutio
n
CGCC130MSD (A11)
CGCC130MSD (A12)
STRESSCHECK (A11)
STRESSCHECK (A12)
a11 merged with left edge
a32 merged with a41 and a41 merged with a51
�-solutions for the lead crack a0 (< 50mm)
�-solutions for the lead crack a0(50mm<a0<300mm)
13
CC-130 Global and Local FE Modeling
Full aircraftCenter wing
Lower panel (βas2)
Local model (βas1)
ββββ-solution for adjacent
structural effect and MED
ββββas= ββββas1 * ββββas2
14
Effect of Load Re-distribution (ββββas2)
• Methodology:– Detach elements in global FEM
• Crack faces• Stringers when failed
– Sum of loads across WS61• skin, cap, stringer
Detailed FEM is needed to refine the results
a = 20 in, no stringer failure
ββββas2
15
Effect of Cap/Stringer and Load Re-distribution
Assumption: Stringer #24 fails when the lead crack reach 12-inch; stringer #23 fails at 17-inch
9) .( 2
1
21
FigreductionLoadas�
uKsK
�aTu�
�aTs�as�
as�as�as�
=
==
∗=
16
Crack Growth Analysis Tool
• CGA Software: NRC Crack Growth Software, CGCC130MSD– β-library (or user defined β)– Standard crack problem (single dominant crack, phase-by-phase )– MSD problem– Forman Equation and Retardation (Hsu model)– Monte Carlo simulation– In-service finding regression
• Spectrum: Medium usage spectrum developed by L3-Spar and used by QETE for coupon testing of CFCW-1
17
SC vs. MSD:ββββ-Solutions
18
SC-PBP
SC
MSD
SC vs. MSD:Life Prediction
~25%
Using NRC Crack Growth Software, CGCC130MSD
OEM DTA Duplicating
19
MSD/MED Residual Strength Analysis
• RS failure criteria used:– Ultimate or yield strength (σult, σys)– Fracture toughness – Abrupt Fracture (Kcr)
Stringer #24 failedStringer #23 failed
Residual strength (normalized to �ys) curves for SC and MSD/MED scenarios
��
�
�
��
�
�=
�a�(a)
K,�(a)RS� C
ysmin
20
ICSD/EIFSD Methodologies
• Approach 1 (ICSD/EIFSD): with a small sample size (n < 40) of crack data from service, full scale test, and/or teardown
• Approach 2 (ICSD/EIFSD): with an extremely small sample size (n<5) of crack data from service or full scale tests
• Approach 3 (IDS/HOLSIP): with no crack data available from service, material and/or coupon test data can be used to determine an ICSD
Affecting Factors
• DaDTA vs DTA curve
• Lognormal vs. Weibull
• Uncensored vs. censored sample
• Confidence bands
• Effect of NDI uncertainty
Ref: RTO-MP-AVT-157 (Montréal, 2008)
21
ICSD/EIFSD Approach 1
• Direct regression in-service findings to EIFS, and then find a best-fit statistical distribution
0.00001
0.0001
0.001
0.01
0.1
1
0 10,000 20,000 30,000 40,000Flight hour
Cra
ck L
engt
h (in
) In-service finding
xxxx
EIFS
For small sample (n<40) crack data from service/full scale test/teardown
Regression (back calculation) methods:
a) Using DaDTA/DTA curve (Master curve) b) Using the calibrated crack growth program
Similar results are obtained using both methods
22
MSD/MED Monte Carlo Simulation
Monte Carlo Random EIFS generator
Crack growth from EIFS
Crack size (a) vs. time (t)
Crack size distribution at time ti ,F(a)
Probability of Failure (PoF)
x N
t
a
t1 t2 t3 t4
START
23
EIFSD for MSD and Monte Carlo F(a)
EIFSD and MSD/MED Monte Carlo crack size distribution F(a) matched in-service findings
24
MSD/MED Monte Carlo Simulation
• Challenge: TIME!– 3.5 min./trial x 1,000,000 trials (Laptop) > 6 years!!!
• Strategy– Reduce number of trials to 100,000: 8 months– 10% tails results (10,000 trials only): 24 days– Parallel computing (NRC’s Linux cluster):
• 24 days / 84 CPU = 7 hours• 24 days / 25 CPU = 23 hours
Still has room to improve!
25N=10,000 of 100,000 runs
Matched
Matched
Crack Size Distribution F(a) Tail Sampling
10% tail sampling
100% sampling
N=10,000 runs
26
Max. Stress Distribution (per flight hour)
Example stress exceedance curves
• Table look-up format was used as they fit the data better than a Gumbel distribution
1E-111E-101E-091E-081E-071E-061E-051E-041E-031E-021E-011E+00
0 0.2 0.4 0.6 0.8 1Maximum stress as a ratio of limit stress
Pro
babi
lity
of e
xcee
danc
es p
er H
our
(1-C
DF
)
Gumbel fit
Table look-updata (CF2004)
Max. stress distribution
MaxMin
27
PoF Results for SC and MSD/MED
• PoF(MSD) is significantly HIGHER than PoF(SC), especially after a certain point in the end of service life
• Maintenance actions should be adjusted according to MSD/MED based crack growth, residual strength, and risk analyses
1E-121E-111E-101E-091E-081E-071E-061E-051E-041E-031E-021E-011E+00
0 10000 20000 30000 40000 50000 60000 70000 80000
EBH
Sin
gle
hour
PoF
, PoF
(t)
Standard crack scenario(Monte Carlo, ProDTA)
MSD scenario (MonteCarlo, ProDTA) ~24%
28
Discussion: Master Curve vs. Monte Carlo
• The master curve and Monte Carlo approaches gave similar PoF. Since the master curve approach is significantly faster than the Monte Carlo approach, further investigation is worthwhile for MSD/MED risk analysis
1E-121E-111E-101E-091E-081E-071E-061E-051E-041E-031E-021E-011E+00
0 10000 20000 30000 40000 50000 60000 70000 80000
EBH
Sin
gle
hour
PoF
, PoF
(t)
ProDTA: SC (Monte Carlo,same EIFSD)
ProDTA: SC (Master curveapproach)
ProDTA: MSD (Mastercurve approach)
ProDTA: MSD (MonteCarlo, same EIFSD)
29
• Developed advanced MSD/MED analysis methodologies and tools, including beta library, crack growth analysis, residual strengthanalysis, and Monte Carlo simulation to support DaDTA of build-up structures of aircraft like CC-130, CP140
• Developed ICSD/EIFSD using CF in-service damage data
• Improved NRC-ProDTA software to calculate the PoF for MSD/MED scenario, using Monte Carlo based MSD crack size distributions
• Results showed that the PoF of MSD is significantly higher than the PoF of SC (standard crack), especially after a certain point in the end of service life. The maintenance actions can be adjusted according to the MSD/MED risk analysis, crack growth, and residual strength analysis results.
Concluding Remarks
30
Future Work: Quantitative Risk Assessment
(Project funded in 2008-2011)• Objectives:
– Continue DaDTA and PoF studies for other locations in CC-130 and CP-140 aircraft
– Support the CF life cycle management • Partners:
– Structures, NDE, DTAES/DND, IMP Aerospace …