1

DATA FUSION APPLIED TO DATA FUSION APPLIED TO HELICOPTER CONDITION HELICOPTER CONDITION

ASSESSMENTASSESSMENT2ND Wind Turbine Reliability Workshop

September 17-18, 2007Albuquerque, New Mexico

Paula DempseyNASA Glenn Research Center

2

OVERVIEWOVERVIEW

Background• Helicopter Transmission Condition Indicators• Health Usage Monitoring System Challenges• Data Fusion

ObjectiveCondition Indicators (CI)

• Gear/Bearing Vibration CI, Oil Debris CIData Fusion Analysis

• Gear/Bearing Examples

Summary

3

BACKROUNDBACKROUND

Helicopter safety is heavily dependent on the reliability and integrity of the mechanical components in the transmission.

Damage in transmission components produce specific fault patterns in vibration signatures.

Vibration-based condition indicators (CI) are used to assess transmission health and diagnose component damage.

Various vibration signature analysis methods developed to detect damage to bearings, gears, etc.

Appropriate vibration analysis techniques must be use for the range of fault detection capability required.

4

BACKROUNDBACKROUND

Oil analysis of wear debris is also used to indicate abnormal wear related conditions in the transmission

Gear/bearing fatigue failures produce significant wear debris in oil lubrication systems.

Oil debris monitoring for gearboxes: off-line oil analysis, plug type chip detectors, in-line inductance type sensors.

Tradeoffs between CI threshold sensitivity to damage and false alarms.

5

BACKROUNDBACKROUND

Commercial Helicopter HUMS (Health Usage Monitoring Systems) providing safety and economic benefits.

Challenges that still need addressed:Increase the fault detection coverage from today’s rate of 70 %Decrease false alarm rates Develop technology to assess severity of damage magnitudeDevelop life prediction technologies Integrate health monitoring outputs with maintenance actionsProcess diagnostic information with minimal expert interpretationCorrelate failure under seeded fault tests with operational dataIn-flight pilot cueing/warning of catastrophic failures

6

Fault DetectionFault Detection

•• Improve performance of CI for Improve performance of CI for component/fault typecomponent/fault type

•• Establish CI performance standardsEstablish CI performance standards•• Establish thresholds with limited fault Establish thresholds with limited fault

datadata•• Develop Health Indicators by Develop Health Indicators by

fusing CI from multiple sensorsfusing CI from multiple sensors

Damage Damage AssessmentAssessment

Remaining LifeRemaining Life

•• Identify damaged componentIdentify damaged component•• Define damage magnitude Define damage magnitude

(correlate with fault detection CI(correlate with fault detection CI

•• Develop life modelsDevelop life models•• Develop life databaseDevelop life databaseDiagnostics

Prognostics

BACKROUNDBACKROUND

7

Data fusion: Data from multiple sensors are combined to perform inferences that are not possible from a single sensor.

Data fusion analysis has been applied to vibration and oil debris condition indicators.

Similar to how we integrate data from multiple sources and senses to make decisions.

Sensor data can be fused at the raw data level, CI level, or decision level.

Decision level fusion the most flexible, does not limit the fusion process to a specific CI.

BACKROUNDBACKROUND

8

OBJECTIVEOBJECTIVE

Provide examples of multi-sensor data fusion applied to vibration and oil debris CI to detect mechanical component health.

Demonstrate integration of two measurement technologies, oil debris analysis and vibration, provide clear information to the end user on the health of the transmission.

9

GEAR VIBRATION CIGEAR VIBRATION CI

Traditional methods are based on the statistical measurement of vibration energy.

Vibration signature analysis methods utilize synchronous averaging. - Gears produce vibration signals synchronous with speed

Time synchronous averaging -Divide signal into 1 revolution segments, then average

Reinforces vibration periodic with shaft speed.

10

Time synchronous averaged (TSA)vibration data

Fast Fourier Transform

Filter

Inverse Fast Fourier Transform

Difference Signal(Filtered TSA) = d

FM4 - Gear Meshing Frequency Harmonics, 1st Order Sidebands

GEAR VIBRATON CIGEAR VIBRATON CI

( )2

1

2

1

4

)(1

1

⎥⎦

⎤⎢⎣

⎡−

−

∑

∑

=

=

N

nn

N

nn

ddN

ddN

INPUTINPUTSensorSensorDataData

OUTPUT OUTPUT CI CI

FM42

3

4

5

6

7

8

9

0 5000 10000 15000 20000

Reading

FM4

11

Fault/defect frequencies generated when bearing fails.

Calculated by bearing dimensions and speed.

Frequency Domain:• FFT used to identify characteristic bearing defect frequencies

and their change in amplitude.• Envelope analysis used to identify bearing resonances excited

by periodic impacts (correlate to defect frequencies) when defect contacts another bearing surface.

Time domain: • Statistical parameters: RMS, peak, kurtosis.

BEARING CIBEARING CI

12

OIL DEBRIS CIOIL DEBRIS CI• Inductance type, in-line, full flow, oil debris sensor• Particle causes disturbance to magnetic field• Measures number, size and accumulated mass of particles• Size (150-1016 micron diam.) distributed in 16 bins• CI – mass or counts

0

20

40

60

80

100

120

140

160

0 1000 2000 3000 4000 5000

Reading

Oil

Deb

ris M

ass

(mg)

Spiral-bevel pinion spallSpiral-bevel pinion spallPinion bearing spallPinion bearing spall

0

20

40

60

80

100

120

250

300

350

400

450

500

550

600

650

700

750

800

863

958

Average Particle Diameter

No.

of P

artic

les

PinionBearing

13

Vibrations measured by the sensor are dominated by the meshing of the planet gear with the ring gear and the sun gear.

Methods under development for separating vibration signatures ofindividual sun, planet, and ring gears.

PLANETARY GEAR CIPLANETARY GEAR CI

Planet 1 Planet 2 Planet 3

Vibration Signal Over One Carrier Rotation

Time

Planet 3

Planet 2

Carrier Directionof Rotation

Planet 1Accelerometer

SunRing

14

DATA FUSION ANALYSISDATA FUSION ANALYSIS

• Data fusion used to integrate vibration and oil debris CI• Data fusion performed at the decision fusion level to determine

gear health.• Fuzzy logic used to identify damage level and integrate CI• Fuzzy logic is used to translate imprecise knowledge into a rule

based system. • Extends Boolean logic to handle partial truths (continuous values 0

to 1).• Data belongs to a set based on its degree of membership.• Fuzzy membership functions defined by analysis of CI and

damage level• Rules defined by data analysis and experts.• Decision level fusion integrates membership functions with rules. • Health of the mechanical component (O.K., Inspect, Shutdown) is

the output.

15

Damage Level FusionMembership FunctionsSensors

Accels1X/RevOil Debris

DATA FUSION ANALYSISDATA FUSION ANALYSIS

Rule FM4 NA4 Debris Output 1 DL DL DL O.K. 2 DH DH DH Shutdown3 DL DL DM Inspect 4 DL DH DL O.K. 5 DL DL DH Inspect 6 DH DL DL O.K. 7 DH DL DM Inspect 8 DH DH DL Inspect 9 DH DL DH Shutdown

10 DH DH DM Inspect 11 DL DH DH Shutdown12 DL DH DM Inspect

0

0.25

0.5

0.75

1

1.25

0 0.25 0.5 0.75 1

Gear State

Mem

bers

hip

O.K. Inspect Shutdown

CI’s –InputsFM4, NA4, Debris

(mg)

Rules Damage Levels

Decision

Gear Health

O.K.InspectShutdown

16

FAILURE MODE: PITTING FATIGUEFAILURE MODE: PITTING FATIGUE

Destructive Pitting: Pits > 0.4 mm and > 25% tooth area

Spiral Bevel Gear Pinion Spur Gear

17

GEAR EXAMPLESGEAR EXAMPLES

• Wear debris and vibration signatures generated during failures.

• Data fusion concept validated in ground tests on Spur/Spiral Bevel Gears.

• Oil debris membership function rescaled to correlate gear tooth surface contact area.

• Fusing oil debris and vibration CI improved damage detection and decision-making.

• Unanticipated bearing failure reinforces importance of data fusion

FM4 NA4 Debris

Spur Gear Spiral Bevel Gear

18

BEARING EXAMPLEBEARING EXAMPLE

0

0.1

0.2

0.3

0.4

2/3/05 2/8/05 2/13/05 2/18/05 2/23/05 2/28/05 3/5/05

Vibr

atio

n R

MS

0

1000

2000

3000

4000

Oil

Deb

ris M

ass

(mg)

1 2

Axial AccelVertical AccelHoriz AccelOil debris Mass

0

0.2

0.4

0.6

0.8

1

2/3/05 2/8/05 2/13/05 2/18/05 2/23/05 2/28/05 3/5/05

Mod

el O

utpu

t

21

O.K.

Inspect

Shutdown

Tapered Roller Bearing Cone Damage

Cone (inner ring)

19

SUMMARYSUMMARY

Improved diagnostic tool damage detection and decision-making performance using fused system over individual CI

Validation of the diagnostic tools is required to diagnose damage severity and component remaining life.

Strengths, weaknesses and constraints of each measurement technology must be identified to capitalize on CI strengths.

Knowledge can be incorporated into the system relieving the end user of interpreting large amounts of sensor data.

20

ADDITIONAL ADDITIONAL

INFORMATIONINFORMATION

21

OIL DEBRIS ANALYSIS OIL DEBRIS ANALYSIS

0

5

10

15

20

25

30

35

40

0 2000 4000 6000 8000 10000 12000 14000 16000Reading Number

Oil

Debr

is M

ass

(mg)

6

1

2

3

4

5

0

0.25

0.5

0.75

1

0 5 10 15 20 25 30 35 40

ODM Mass (mg)

Dam

age

Leve

l

Oil Debris Mass For Levels of Damage

Output of Fuzzy Logic Model

Benefits• Eliminates the need for an expert diagnostician• Logic built-in to minimize operational effects

22

MEMBERSHIP FUNCTIONSMEMBERSHIP FUNCTIONS

0

0.25

0.5

0.75

1

1.25

0 5 10 15 20 25 30 35 40

Oil Debris Mass (mg)

Mem

bers

hip

DL DM DH

Damage Low (DL)Damage Medium (DM)Damage High (DH)

23

Rule 9 for 1 Reading:

If FM4 is DH and NA4 is DL and Debris is DH then Shutdown

MEMBERSHIP FUNCTIONSMEMBERSHIP FUNCTIONS