Model Based Inference for Wire Chafe Diagnostics...Introduction & Motivation Characterization of...
Transcript of Model Based Inference for Wire Chafe Diagnostics...Introduction & Motivation Characterization of...
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
Model Based Inference for Wire ChafeDiagnostics
S. Schuet K. Wheeler D. Timucin M. KowalskiP. Wysocki
Intelligent Systems DivisionNASA Ames Research Center
Moffett Field, California
Aging Aircraft 2009
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
Team Members and Contact Information
Name email
Kevin WheelerTeam Lead
kevin.r.wheeler at nasa.gov
Dogan Timucin dogan.a.timucin at nasa.gov
Phil Wysocki philip.f.wysocki at nasa.gov
Stefan Schuet stefan.r.schuet at nasa.gov
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
Agenda
1 Introduction & MotivationImportance of Wiring FaultsChafing Faults
2 Characterization of Chafing Damage3D Commercial EM SimulatorExperimental MethodsFault Library
3 Model Based InferenceBayesian FrameworkChafe Model
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
Importance of Wiring FaultsChafing Faults
Recent EWIS Incidents
February 2009 A fire breaks out on-board an A340 VirginAtlantic flight en route from Heathrow to Chicago,which could not be extinguished until after theplane landed and depowered. Investigators laterdiscovered problems with the electrical wiring in abar unit of the plane that was specifically adaptedfor the airline.
January 2008 American Airlines B757 Flight 1738 experiencedsmoking in the cockpit caused by arcing within thewindshield heat system resulting in the cockpitwindshield shattering during the emergencylanding.
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
Importance of Wiring FaultsChafing Faults
Chafing is a dominant EWIS failure mechanismA frequently occurring type of wiring fault
FAA reports 55% in one study†
Navy reports 37% in another study†
Precursors to more significant problems:open and short circuits (cause instrument failure)arcing (causes smoking, fires, or worse!)
†See K. Wheeler et. al., “Aging Aircraft Wiring Fault Detection Survey” for an
overview of these studies and more
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
Importance of Wiring FaultsChafing Faults
Detection of Chafing Damage
Efforts in hardware development for chafe detection havebeen reported, however, little attention has been focusedon detectability and uncertainty
Initial investigations on detectability suggest that chafingon unshielded (e.g., power cables) wires is difficult if notimpossible to detect†
Shielded wire (e.g., high-speed communication cables),however, may generate a detectable signature.
†“The Invisible Fray: A Critical Analysis of the Use of Reflectometry for Fray
Location,” Griffiths et al., IEEE Sensors Journal, vol. 6, no. 3, June 2006.
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
Importance of Wiring FaultsChafing Faults
Chafing in Shielded Electrical Cabling
Chafe first ablates outer insulation, then shield, leavinginner conductors intact
Figure: Chafe progression: 2k, 4k, and 8k cycles beyond short
Time domain reflectometry (TDR) between activeconductors and the shield is proposed for chafe detection
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
3D Commercial EM SimulatorExperimental MethodsFault Library
3D Commercial Electromagnetic Simulator
Computer Simulation Technology (CST)’s MicrowaveStudio is usedWire Types:
Coaxial CableTwisted Shielded Pair
Fault Types:RectangularElliptical (pictured)Multiple Faults
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
3D Commercial EM SimulatorExperimental MethodsFault Library
3D Commercial EM Simulator: Representative Data
Response of twisted shielded pair (TSP) to TDRinterrogation
0 0.2 0.4 0.6 0.8 1−0.1
−0.09
−0.08
−0.07
−0.06
−0.05
−0.04
−0.03
−0.02
−0.01
0
Time (ns)
TD
R R
espo
nse
(V)
2mm x 2mm
2mm x 4mm
2mm x 6mm
2mm x 8mm
2mm x 10mm
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
3D Commercial EM SimulatorExperimental MethodsFault Library
Experimental Methods for Chafe CharacterizationThere are two fundamental modes of chafing damage
Wire movement versus stationary object (e.g., wire rubbingon a bulkhead)Wire movement versus wire movement (e.g., wires in abundle abrating each other)
Two machines have been developed to mimic thesedamage mechanisms
Stationary rod abrasion machineWire-on-wire abrasion machine
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
3D Commercial EM SimulatorExperimental MethodsFault Library
Stationary Rod Abrasion Machine
Specifications Range Optimal SettingStroke Length 1 - 3 cm 1 cm
Stroke Frequency 1-100 Hz 10 Hz
200g mass used to pressure wire against diamond coated chafing rod
Average chafe to inner conductor time for TSP is ∼ 12k cycles
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
3D Commercial EM SimulatorExperimental MethodsFault Library
Wire-on-Wire Abrasion Machine†
Specifications Range Optimal SettingStroke Length 1 cm 1 cm
Stroke Frequency 1-20 Hz 10 Hz
500g mass used to pressure wire against diamond coated chafing rod
Average chafe to inner conductor time for TSP is ∼ 25k cycles
†Thanks to AFRL for the design
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
3D Commercial EM SimulatorExperimental MethodsFault Library
Rod Abrasion Machine: Representative DataTwo fault example, one fault fixed and the other growing in size
51 52 53 54 55 56 57 58
0.136
0.138
0.14
0.142
0.144
0.146
0.148
Time (ns)
TD
R R
espo
nse
(V)
(12670, 0) cycles
(12670, 1250) cycles
(12670, 3250) cycles
(12670, 5250) cycles
(12670, 7250) cycles
(12670, 9250) cycles
(12670, 11250) cycles Fixed Fault
Growing Fault
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
3D Commercial EM SimulatorExperimental MethodsFault Library
Wire-on-Wire Abrasion Machine: Representative DataGrowing braid on wire TDR data set
6 7 8 9 10 11 12 13 14 15 16
0.124
0.126
0.128
0.13
0.132
0.134
Time (ns)
TD
R R
espo
nse
(V)
baseline
4000 cycles
6000 cycles
8000 cycles
10000 cycles
12000 cycles
14000 cycles
16000 cycles
Growing Fault
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
3D Commercial EM SimulatorExperimental MethodsFault Library
Overview of Fault LibraryContains TDR response signals from both simulations andlab experiments
Formatted in ASCII and Matlab binary files (.mat)Simulated data was collected by growing fault length andsimulating the TDR responseExperimental data was collected by growing faults undercontrolled chafing conditions and measuring the TDRresponse
Library Available Online:
http://ti.arc.nasa.gov/project/wiring/
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
Bayesian FrameworkChafe Model
Why Use Probability Theory for Wire Fault Detection?Want to infer variables of interest from noisy reflectedelectrical signals:
fault location(s)fault size(s)
Want to automatically cope with sources of uncertainty:electrical noise from equipment and environmentunknown or uncertain cable parameters (e.g., dielectricpermittivity, finite conductivity)geometric distortions (e.g., bends, wiggles)other reflection sources (e.g., splices)unknown number of faults all mixing together
Specifically, the Bayesian approach provides a systematicapproach to incorporating these effects and more...
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
Bayesian FrameworkChafe Model
Benefits to the Bayesian Approach
Clearly represents uncertainty inherent withinmeasurementsIncludes uncertainty in known prior information orexpertise
e.g.,“known”’ values, such as permittivity of wire insulation,are often better represented as random variables knownonly to a certain accuracy
Quantifies the uncertainty in the inferred parameters
Avoids taking direct inverse in finding optimal modelparameters by seeking the estimate that maximizes theprobability of the observed information
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
Bayesian FrameworkChafe Model
Conducting Research in two areas:
Forward modeldevelopment:
LTI Convolution ModelsAnalytical modelsBehavioral models (howthings change)
Optimization Techniques to retrieve parameters (find mostlikely parameters that explain the observed input andoutput).
Convex Optimization & Expectation MaximizationMarkov Chain Monte-Carlo (MCMC)Reversible Jump MCMC
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
Bayesian FrameworkChafe Model
Example: LTI System Model
TDR Θ0 Θ1 Θ2 . . . ΘN−1
ZL
Vi (k)
Vr (k)
Vr (k) = Θ0Vi(k) + Θ1Vi(k − 1) + ... + ΘN−1Vi(k − N + 1)
The reflection coefficients Θk and input Vi(k) are given
Motivated through physics by assuming the line is losslessand linear time invariant (LTI)
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
Bayesian FrameworkChafe Model
Example LTI System Model
TDR Θ0 Θ1 Θ2 . . . ΘN−1
ZL
Vi (k)
Vr (k)
Vr (k) = Θ ∗ Vi (k)
Θ ∈ RN is the variable we want to estimateF (Θ) = Θ ∗ Vi = HΘ represents our model
H ∈ RN×N , is a convolution matrix
y = F (Θ) + ν, where ν ∈ RN is Gaussian noise
Prior information is that Θ is sparse, since chafing damageis small and localized
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
Bayesian FrameworkChafe Model
Likelihood: Prob (y |F ,Θ) ∝ e− 12σ
2 ‖F (Θ)−y‖2
Prior: Prob (Θ|F ) ∝ e−PN−1
k=0 λk |Θk |
A heuristic for prior information that Θ is sparse
Solve: maximize Prob (Θ|F , y) ∝ Prob (y |F ,Θ) Prob (Θ|F ),which is equivalent to,
minimize1
2σ2 ‖F (Θ) − y‖2 +
N−1∑
k=0
λk |Θk | (1)
For fixed λk , (1) is a convex optimization problem, and thussolvable globally and efficiently, even for large N.
The Expectation Maximization algorithm can be used toautomatically find the best “tuning” parameters λk .
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
Bayesian FrameworkChafe Model
Example LTI Convolution Model Estimation ResultN = 1024, ∆t = 0.04 ns
0 5 10 150
0.02
0.04
0.06
0.08
0.1
0.12
0.14
0.16
0.18
0.2
input signalreflected signal with faultreflected signal without fault
6 6.5 7 7.5 80.12
0.13
0.14
chaffing damage
TDR Data
V(k
∆t)
k∆t (ns)
0 5 10 15−0.05
−0.04
−0.03
−0.02
−0.01
0
0.01
0.02
0.03
0.04
0.05Fault Detection
Θ with faultΘ without fault
6 6.5 7 7.5 80
10
20x 10
−4Θk
k∆t (ns)
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
Bayesian FrameworkChafe Model
Mathematical Predictive Chafe Model: Impedance Layering
Assume that chafe can be thought of as one or a series ofimpedance disturbances
Once impedance of fault is known, it is relativelystraightforward to find the response of the cable infrequency or time domains
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
Bayesian FrameworkChafe Model
Computing Capacitance and Inductance
Capacitance:
∇ · ǫ∇φ = 0Ql =
RR
∇ · D ds=
H
−ǫ∇φ · (n × z) dl
Inductance:
Ll =µ0ǫ
Cl
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
Bayesian FrameworkChafe Model
Frequency Domain Verification : Comparison with CST MWS
Predicted Return Loss from a 2 × 10 mm chafe (left) and a2 × 15 mm chafe (right) in coaxial cable.
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
Bayesian FrameworkChafe Model
Experimental Verification: Chafe in Coaxial Cable
NASA Ames Research Center Wire Chafe Diagnostics
Introduction & MotivationCharacterization of Chafing Damage
Model Based Inference
Bayesian FrameworkChafe Model
Experimental Verification: Comparison to Lab TDR
NASA Ames Research Center Wire Chafe Diagnostics
ConclusionReferences
Conclusions
Machines – Developed two chafing machines used tomimic effect of chafing on cables.
Datasets – Development of publicly accessible electricalsignature fault datasets.Algorithms:
Development of probabilistic Bayesian algorithms forunderstanding and characterizing electrical signatures offaultsDevelopment of compact and efficient electromagneticsbased forward models for chafe signature
NASA Ames Research Center Wire Chafe Diagnostics
ConclusionReferences
Future Work
Incorporation of forward models within a Bayesianframework is underway
“Real-life” experimental platforms (NASA wind tunnel andvertical motion simulator) are being used
Live communication cable interrogation (CAN bus) is beingmodeled and contemplated as representative platform.
NASA Ames Research Center Wire Chafe Diagnostics
ConclusionReferences
References
K. Wheeler et al.Aging Aircraft Wiring Fault Detection Surveyhttp://ti.arc.nasa.gov/m/pub/1342h/1342 (Wheeler).pdf
D. SiviaData Analysis, A Bayesian TutorialOxford Science Publications, 2003
S. Boyd, L. VandenbergheConvex OptimizationCambridge University Press, 2003http://www.stanford.edu/∼boyd/cvxbook/
NASA Ames Research Center Wire Chafe Diagnostics