The document discusses using a particle filter approach for fault tolerant navigation of an ROV. It describes the ROV system and sensors, including common faults like dropout of the hydro-acoustic position reference and bias in the Doppler velocity log. It explains how a particle filter can be used to both estimate the ROV's state and detect faults by comparing predictions under fault-free and faulty models to sensor measurements. Simulation results demonstrate the effectiveness of this integrated particle filter for navigation and fault detection. The approach provides a straightforward modeling framework that can perform navigation and fault handling within a single structure and is extensible to new fault modes.

1. Fault Tolerant ROV Navigation System
based on Particle Filter
using Hydro-acoustic Position
and Doppler Velocity Measurements

2. Bo Zhao, Ph.D. candidate in CeSOS, NTNU
Research topic: Fault tolerant control for DP
?-2009 M.Eng. in Navigation, Guidance and Control (for aircrafts)
Nov. 2009 Start my Ph.D.
Spring, 2010 Courses, preliminary research
Fall, 2010 Courses, preliminary research
Spring, 2011 Courses in DTU, Denmark. Hooked up with the particle filter
Fall, 2011 Course, research, and papers
Spring, 2012 Research, papers, go to conferences, prepare for experiment
Fall, 2012 Research, papers, go to conferences, do experiment

3. 2. System modeling
1. Introduction
3. Fault analysis and
modeling
5. Results
4. Particle filter for fault
detection

4. y
x
z

5. y
x
z
Width: 82 cm
Height: 80 cm
Length: 144 cm
Net weight: 405 kg
Payload: 20 kg

6. y
x
z
2×Vertical thrusters
Vertical: 1.2 knot
2×Main thrusters
Tunnel thruster
Yaw rate: 60°/s

7. y
x
z
Lights
Camera
Manipulators

8. y
x
compass
Yaw rate gyro z
HPR (Hydroacoustic
position reference)
DVL (Dopple Velocity Log)
depth sensor

9. HPR
– Hydro acoustic position reference
Faults:
1. Dropout – when no signal received
2. Outlier – Measurement has
significant difference from the true
position

10. DVL
– Doppler velocity log
Faults:
1. Dropout – when no signal received
2. Bias – small-size constant difference
between the measurement and the
true velocity

11. Navigation: Obtain the position and velocity of the ROV
Disturbance and noise
1. System noise
2. Model uncertainty
3. Measurement noise
4. Current
5. Failures

12. Navigation: Obtain the position and velocity of the ROV
Disturbance and noise Failure modes
1. System noise 1. HPR dropout
2. Model uncertainty 2. HPR outlier
3. Measurement noise 3. DVL dropout
4. Current 4. DVL bias
5. Failures 5. Thruster loss

13. 2. System modeling
1. Introduction
3. Fault analysis and
modeling
5. Results
4. Particle filter for fault
detection

15. Observer for ROV :
Particle filter
Pictures from
http://www.gris.uni-tuebingen.de/people/staff/sfleck/smartsurv3d/
http://perception.inrialpes.fr/~chari/myweb/Research/
http://wires.wiley.com/WileyCDA/WiresArticle/articles.html?doi=10.1002%2Fwics.1210

16. 2. System modeling
1. Introduction
3. Fault analysis and
modeling
5. Results
4. Particle filter for fault
detection

17. Failure modes
1. HPR dropout
2. HPR outlier
3. DVL dropout
4. DVL bias
5. Thruster loss

18. HPR data
HPR update interval
Failure modes
1. HPR dropout
2. HPR outlier
3. DVL dropout
4. DVL bias
5. Thruster loss

19. HPR data
HPR update interval
Failure modes
1. HPR dropout
2. HPR outlier
3. DVL dropout
4. DVL bias
5. Thruster loss

20. 0
-5
-10
East velocity [m/sec]
DVL data -15
-20
-25
-30
-35
5400 5600 5800 6000 6200
Time [sec]
Failure modes
1. HPR dropout
2. HPR outlier
3. DVL dropout
4. DVL bias
5. Thruster loss

21. Failure modes
1. HPR dropout
2. HPR outlier
3. DVL dropout
4. DVL bias
5. Thruster loss

22. Failure modes
1. HPR dropout
2. HPR outlier
3. DVL dropout
4. DVL bias
5. Thruster loss

23. Comment: 0
-5
0. If the fault in the system is known, we can
-10
m/sec]
design an filter to solve the observation problem -15
locity [
1. It is not easy to design observers for the -20
East ve
system models in different failure modes-25
2. Even if a bank of observers is designed, it is
-30
hard to decide which one to use, since the
-35
5400
failure mode is unknown. 5600
5800
Tim e 6000
[se c]
6200
Failure modes
1. HPR dropout
2. HPR outlier
3. DVL dropout
4. DVL bias
5. Thruster loss

24. 2. System modeling
1. Introduction
3. Fault analysis and
modeling
5. Results
4. Particle filter for fault
detection

25. How do we cognize the world?
Observation
Prediction Correction

26. How do we diagnose a fault?
Prediction
Predicted
Fault free behavior
Predicted
Faulty behavior

27. How do we diagnose a fault?
Prediction
Predicted
Fault free behavior
Predicted
Faulty behavior

28. How do we diagnose a fault?
Prediction Observation
Predicted Take the measurement
Fault free behavior
Correction
Obs
H1
Predicted Compare
Faulty behavior H2

29. Introduction to Particle Filter
Outline
System States
State Estimation
Kalman Filter
Particle Filter
Case Study
p

30. Introduction to Particle Filter
Outline
System States
State Estimation
Kalman Filter Measuring
Particle Filter pm
Case Study
p

31. Introduction to Particle Filter
Outline
System States
State Estimation
Kalman Filter Estimating
Particle Filter
Case Study
pm

32. Introduction to Particle Filter
Outline
System States
State Estimation
Kalman Filter Estimating
Particle Filter
p
Case Study
pm

33. Correction Obs
H1
H2
p
pm

34. How do we diagnose a fault?
Prediction Observation
Predicted Take the measurement
Fault free behavior
Correction
Obs
H1
Predicted Compare
Faulty behavior H2

35. 2. System modeling
1. Introduction
3. Fault analysis and
modeling
5. Results
4. Particle filter for fault
detection

36. What has been talked about?
• ROV, and its navigation sensors
• Faults in the sensors and their model
• The concept of fault detection with
particle filter
• Simulation results
What are the advantages?
• Straight-forward modeling
• Do the navigation and fault
handling with in a single structure
• Extendable