F.L. Lewis, Assoc. Director for ResearchMoncrief-O’Donnell Endowed Chair
Head, Controls, Sensors, MEMS GroupAutomation & Robotics Research Institute (ARRI)
The University of Texas at Arlington
Automation & Robotics Research Institute (ARRI)The University of Texas at Arlington
F.L. Lewis, Assoc. Director for ResearchMoncrief-O’Donnell Endowed Chair
Head, Controls, Sensors, MEMS Group
http://ARRI.uta.edu/acs
CBM II - Diagnostics
ObjectivesExtend equipment lifetimeReduce down timeKeep throughput and due dates on trackUse minimum of maintenance personnelMaximum uptime for minimum effective maintenance costsCBM should be transparent to the user
No extra maintenance for the CBM network!Determine the best time to do maintenance
Efficiently use maintenance & repair resourcesDo not interfere with machine usage requirements
Allow planning for maintenance costsNo unexpected last-minute costs!
Condition-Based Maintenance (CBM)Prognostics & Health Management (PHM)
The CBM/PHM Cycle
MachineSensors
Pre-Processing
FeatureExtraction
FaultClassification
Predictionof Fault
EvolutionData
ScheduleRequired
Maintenance
Systems &Signal processing
Diagnostics Prognostics MaintenanceScheduling
Identify importantfeatures
Fault Mode Analysis
Machine legacy failure data
Available resourcesRULMission due dates
Required Background Studies
Contact George [email protected]://icsl.marc.gatech.edu
PHMCBM
Off Line- Background Studies, Fault Mode AnalysisOn Line- Perform real-time Fault Monitoring & Diagnosis
Two Phases of CBM Diagnostics
Three Stages of CBM/PHM
DiagnosticsPrognosticsMaintenance Scheduling
CBM – Fault Diagnosis Background Studies
• Fault Mode Analysis (FMA) - Identify Failure and Fault Modes
• Identify the best Features to track for effective diagnosis
• Identify measured sensor outputs needed to compute the features
• Build Fault Pattern Library
Deal with FAULTSNeed to identify Faults before they become Failures
Phase I- Preliminary Off Line Studies
Compressor Pre-rotation Vane
Condenser
Evaporator
•Compressor Stall & Surge•Shaft Seal Leakage•Oil Level High/Low•Aux. Pump Fail•Oil Cooler Fail•PRV/VGD Mechanical Failure
•Condenser Tube Fouling•Condenser Water Control Valve Failure•Tube Leakage•Decreased Sea Water Flow
•Target Flow Meter Failure•Decreased Chilled Water Flow•Evaporator Tube Freezing
•Non Condensable Gas in Refrigerant•Contaminated Refrigerant•Refrigerant Charge High•Refrigerant Charge Low
•SW in/out temp.•SW flow•Cond. press.•Cond. PD press.•Cond. liquid out temp.
•Comp. suct. press./temp.•Comp. disch. press./temp.•Comp. oil press./flow (at required points)•Comp. bearing oil temp•Comp. suct. super-heat•Shaft seal interface temp.•PRV Position
•Liquid line temp.•(Refrigerant weight)
•CW in/out temp./flow•Eva. temp./press.•Eva. PD press.
Ex. Ex. -- Navy Centrifugal Chiller Failure ModesNavy Centrifugal Chiller Failure Modes
Fault Mode Analysis Contact George [email protected]://icsl.marc.gatech.edu
Fault Modes of an Electro-Hydraulic Flight Actuator
V. Skormin, 1994SUNY Binghamton
bearingcontrol surface
hydrauliccylinder
pump
poweramplifier
Fault Modes
Control surface lossExcessive bearing friction
Hydraulic system leakageAir in hydraulic systemExcessive cylinder frictionMalfunction of pump control valve
Rotor mechanical damageMotor magnetism loss
motor
Fault Mode Analysis
Use Physics of Failure and Failure Models to select failure features to include in feature vectors
Select Fault ID Feature Vector
Method 1- Dynamical System Diagnostic Models
The Fault Feature Vector is a sufficient statistic for identifying existing fault modes and conditions
BJssTs
+=
1)()(ωmotor dynamics
sBsMsFsX
pp )(1
)()(
+=pump/piston dynamics
LsKAsR
sP
+=
)(
1)()(
2actuator system dynamics
Physical parameters are J, B, Mp, Bp, K, L
V. Skormin, 1994SUNY Binghamton
Select Feature VectorRelate physical parameters J, B, Mp, Bp, K, L to fault modes
Get expert opinion (from manufacturer or from user group)Get actual fault/failure legacy data from recorded machine historiesOr run system testbed under induced faults
Result -
Etc.Etc.
THEN (fault is air in hydraulic system)IF (actuator stiffness K is small)AND (piston damping coeff. Bp is small)
THEN (fault is excess cylinder friction)IF (motor damping coeff. B is large)AND (piston damping coeff. Bp is large)
THEN (fault is hydraulic system leakage)IF (leakage coeff. L is large)Fault ModeCondition
Therefore, select the physical parameters as the feature vectorT
pp LKBMBJt ][)( =φ
V. Skormin, 1994SUNY Binghamton
Select Sensors for the Best Outputs to Measure
V. Skormin, 1994SUNY Binghamton
Tpp LKBMBJt ][)( =φ
Cannot directly measure the feature vector
Can measure the inputs and outputs of the dynamical blocks, e.g.
BJssTs
+=
1)()(ω
)(2
)()( tPDtCItTπ
−= ω(t)motor speed
armaturecurrent I(t)
pressuredifference P(t)
Therefore, use system identification techniques to estimate the features
Virtual Sensors = physical sensors + signal processing se
nsor
sDSP
signals from machine
Fault IDfeatures
Method 2- Non-Model-Based Techniques
Select Fault ID Feature Vector
Etc.Etc.
THEN (fault is worn outer ball bearing)IF (third harmonic of shaft speed is present)AND (kurtosis of load vibration is large)
THEN (fault is gear tooth wear)IF (shaft vibration second mode is large)AND (motor vibration RMS value is large)
THEN (fault is unbalance)IF (base mount vibration energy is large)Fault ModeCondition
Therefore, include vibration moments and frequencies in the feature vector
=)(tφ [ time signals … frequency signals ]T
Get expert opinion (from manufacturer or from user group)Get actual fault/failure legacy data from recorded machine historiesOr run system testbed under induced faults
Method 3- Statistical Regression Techniques
Select Fault ID Feature Vector
Vibration magnitude
Driv
e tra
in g
ear t
ooth
wea
r
Pearson’s correlationNonlinear correlation techniquesMultivariable regression
Clustering techniquesNeural networksStatistical
Fault 1
Fault 2
Fault 3
outliers
Etc.Etc.
THEN (fault is air in hydraulic system)IF (actuator stiffness K is small)AND (piston damping coeff. Bp is small)
THEN (fault is excess cylinder friction)IF (motor damping coeff. B is large)AND (piston damping coeff. Bp is large)
THEN (fault is hydraulic system leakage)IF (leakage coeff. L is large)Fault ModeCondition
Fault Pattern Library
Etc.Etc.
THEN (fault is worn outer ball bearing)IF (third harmonic of shaft speed is present)AND (kurtosis of load vibration is large)
THEN (fault is gear tooth wear)IF (shaft vibration second mode is large)AND (motor vibration RMS value is large)
THEN (fault is unbalance)IF (base mount vibration energy is large)Fault ModeCondition
Fuzzy Logic Rulebase !
www.MIMOSA.orgMachine User Group- CBM Data
Condition Monitoring and Diagnostics of Machines
CBM Fault DIAGNOSTICS Procedure
machines
Math models
),,(),,(
ππ
uxhyuxfx
==&
System Identification-Kalman filterNN system ID
RLS, LSE
Dig. Signal Processing
PhysicalParameterestimates &Aero. coeff.estimates
π̂
Sensoroutputs
VibrationMoments, FFT
FeatureVectors-
Sufficientstatistics
)(tφFault ClassificationFeature patterns for faultsDecision fusion could use:
Fuzzy LogicExpert SystemsNN classifier
Stored Legacy Failure dataStatistics analysis
Feature extraction -determine inputs for Fault Classification
Physics of failureSystem dynamicsPhysical params.
Identify Faults/Failures
Set Decision ThresholdsManuf. variability dataUsage variabilityMission historyMinimize Pr{false alarm}Baseline perf. requirements
More info needed?
Inject probe test signals for refined diagnosisInformpilotyes
π
Serious?
Informpilot
yes
SensingFault Feature Extraction
Reasoning& Diagnosis
Systems, DSP& Data Fusion
SensorFusion
Featurevectors
Featurefusion
StoredFault Pattern
Library
Model-BasedDiagnosis
Phase II- On Line Fault Monitoring and Diagnostics
no
Request Maintenance
Fault Classification
Decision-MakingFault Classification
StoredFault Pattern
Library
Feature Vectors
)(tφ
Diagnosed Faults
Neural networksFuzzy logicExpert system rulebaseBayesianDempster-Shafer
Model-Based Reasoning (MBR) vs. Case-Based Reasoning
Too complex!Faults often depend on Operating conditions
Fuzzy Logic Fault ClassificationUnifies
expert systemsstatisticalneural network approaches
2-D FL system c.f. neural network
Fig 1 FL rulebase to diagnose broken bars in motor drives usingsideband components of vibration signature FFT [Filippetti 2000].
Number of broken bars = none, one, two.Incip. = incipient fault
small medium large
smal
lm
ediu
mla
rge
Sideband component I1
Side
band
com
pone
nt I 2
none incip.
incip.
one
one
one
oneortwo
oneortwo
two
... ..
.........
.................... .
......... . . . ...... .
.. ..
.
. ..
Fig 5 Clustering of statistical fault data
Vibration magnitude
Driv
e tra
in g
ear t
ooth
wea
r
Faul
t con
ditio
ns
one
two
thre
e
low med severe
FL Decision Thresholds
From Chestnut
Based onLegacy fault data historiesManuf. variability dataUsage variabilityMission historyMinimize Pr{false alarm}Baseline perf. requirements
Can be tuned using adaptive NN learning techniques
Mode DiagramExample Testbed: AC-Plant
Contact George [email protected]://icsl.marc.gatech.edu
Possible failures depend on current operating mode
Model-Based ReasoningMBR
Joint Strike Fighter –Prognostics and Health Management (PHM)
‘Joint Strike Fighter’,JSF, and the JSF Logo are Trademarks of the United States Government
Public Released Under: AER200307014
Michael Gandy and Kevin LineLockheed Martin Aeronautics
PHM Architecture and Enabling Technologies
Air Vehicle On-BoardHealth Assessment
Health Management,Reporting & Recording
Autonomic Logistics& Off-Board PHM
PVI
MAINTAINERVEHICLE INTERFACE
Mission Critical
PHMData
Displays & ControlsCrashRecorder
MaintenanceInterface Functions
IETMsConsumables
On-Board Diagnostics
PMD
.
PMA
In-Flight &Maintenance Data Link
Flight Critical
PHM / Service Info
Database
AMD/PMD
PHM Area Managers
MS Subsystems
• Sensor Fusion• Model-Based Reasoning• Tailored Algorithms• Systems Specific
Logic / Rules• Feature Extraction
Provides:
• AV-Level Info Management• Intelligent FI• Prognostics/Trends• Auto. LogisticsEnabling/Interface
Methods Used:
FCS/UtilitySubsystems
NVMICAWSManager
AVPHMHosted in ICP
Structures
MissionSystems
• Decision Support• Troubleshooting and Repair• Condition-Based Maintenance• Efficient Logistics
Vehicle Systems
Propulsion
Results In:
ALIS• Automated Pilot / Maint. Debrief
• Off-Board Prognostics
• Intelligent Help Environment
• Store / Distribute PHM Information
Hosted in ICP
Michael Gandy and Kevin LineLockheed Martin Aeronautics
Model Legend -Model Legend -Condition Function
SensorComponent
BlockDiagram
MBRModel
MBR Approach Provides Multiple Benefits and Functions:– Intuitive, Multi-Level Modeling– Inherent Cross Checking for False Alarm Mitigation– Multi-Level Correlation for Failure Isolation Advantage
Chains of Functions Indicate Functional Flows.– Components Link to the Functions They Support.– Sensors Link to the Functions They Monitor.– Conditions Link to the Functions They Control.
Michael Gandy and Kevin LineLockheed Martin AeronauticsModel-Based Reasoning (MBR) Provides a
Significant Part of PHM Design Solution
PEDS Software System PEDS Software System Architecture Architecture
(Stand(Stand--alonealone))
5. Feature Extraction
5. Feature Extraction4. Feature
Extraction4. Feature Extraction
Case-basedDiagnostic Reasoner
Case-basedDiagnostic Reasoner
Hardware•Plant•Sensors•DAQ
Hardware•Plant•Sensors•DAQ
3. Mode Estimator /Usage PatternIdentification
3. Mode Estimator /Usage PatternIdentification
Central DBEvent DispatchEvent Dispatch
Database ManagementDatabase
Management
CBMmain. schedule FAHPFAHP
Causal Adjustments
Causalfactors
ScenarioGenerator
ScenarioGenerator
DWNN
VirtualSensor(WNN)
CPNN
failuredimension
6.
Classifier(WNN)
Classifier(WNN)
Classifier(Fuzzy)
Classifier(Fuzzy)
5.
2. Data Preprocessing
2. Data Preprocessing
1. GUI1. GUI
Contact George [email protected]://icsl.marc.gatech.edu