Sensor Fusion

Table of Contents
Sensor fusion theory The upgraded testbed Sun sensor Magnetometer Rate gyros Data fusion State fusion Results Shadowed GPS Conclusion

21/12-2000 (MJ)

Danish GPS Center

What is Sensor Fusion?

Combination of different types of sensors to reach a better performance than possible with a single sensor Best sensor fusion system: The human brain

Why use it?

To achieve better performance Extra sensors could work as backup if others fail (redundancy)
21/12-2000 (MJ) Danish GPS Center 3

Complementary Sensors
Example:
FLIR profile

Enemy plane

RADAR profile

Ship

21/12-2000 (MJ)

Danish GPS Center

Fusion Levels
Decision Level

Increasing dataload

Feature Level

Increasing abstraction

Data Level

21/12-2000 (MJ)

Danish GPS Center

Applied Techniques
Fusion Level Data Level Function Spectral Datamining Data Adaptation Estimation of Parameters Feature Level Classification Techniques Digital Signal Processing Coordinate Transforms Unit Adjustments Kalman Filtering Batch Estimation Pattern Recognition Fuzzy Logic Neural Networks Expert Systems Artificial Intelligence

Decision Level

Decide Action

21/12-2000 (MJ)

Danish GPS Center

Decentralized Fusion
Sensor 1 Preprocessing Central fusion processor Sensor 2 Preprocessing Data adaptation

Sensor n

Preprocessing

a) Centralized fusion
Sensor 1 Preprocessing Data adaptation Sensor 2 Preprocessing Central fusion processor Sensor n-1 Sensor n Preprocessing Data adaptation Preprocessing Local fusion processor Local fusion processor

b) Decentralized fusion

21/12-2000 (MJ)

Danish GPS Center

Extra Sensors
Magnetic Field Sunlight GPS signals Angular Rate

Sensor Box

Antenna Array

Solar Sensor Magnetometer Rate Gyros + GPS

Labtop PC

Controller Box

Rotation Stages

Hardware Box

GPS Receiver

RS232

LON network

21/12-2000 (MJ)

Danish GPS Center

The Upgraded Testbed

21/12-2000 (MJ)

Danish GPS Center

Vector Matching
Wahbas Problem:

Solution: Use the SVD method 1. Form B matrix of outer products

21/12-2000 (MJ)

Danish GPS Center

10

Vector Matching
2. Split B into three matrices using Singular Value Decomposition

3. The optimal attitude matrix best fitting the vector pairs can be found as:

21/12-2000 (MJ)

Danish GPS Center

11

Solar Cells
Working Principle:

iss

Model:

21/12-2000 (MJ)

Danish GPS Center

12

Solar Cells
Temperature effects

21/12-2000 (MJ)

Danish GPS Center

13

Solar Cells
Calibration Sensitivity Mounting Cosine char.

21/12-2000 (MJ)

Danish GPS Center

14

Solar Cells
Mounting errors (adjustment of cell-axis)

21/12-2000 (MJ)

Danish GPS Center

15

Solar Cells
Cosine corrections

21/12-2000 (MJ)

Danish GPS Center

16

Sun Sensor
Performance

White Noise?
21/12-2000 (MJ) Danish GPS Center 17

Magnetometer
Calibration Field bias Mounting uncertainties

21/12-2000 (MJ)

Danish GPS Center

18

Magnetometer
Calculating internal bias At zero attitude (A = I)
bias

Each measurement
ref m

Least Square

21/12-2000 (MJ)

Danish GPS Center

19

Magnetometer
Mounting uncertainties calibrated by a small rotation, found using the SVD method Result:

21/12-2000 (MJ)

Danish GPS Center

20

Rate Gyros
Working Principle:
Input fork Oscillator

Amplitude regulator

Connection joint

Ref Amplitude regulator Demodulator

DC voltage

Filter amplifier

Output fork

Measures the state (angular rate) directly

21/12-2000 (MJ) Danish GPS Center 21

Rate Gyros
Error sources: Noise Timevar. bias Scalefactor Accel. sensitivity Off-axis error

21/12-2000 (MJ)

Danish GPS Center

22

Data Fusion
Measurement model:

21/12-2000 (MJ)

Danish GPS Center

23

Data Fusion
Measurement gradient matrix:

21/12-2000 (MJ)

Danish GPS Center

24

State Fusion
Using receiver solution as input measurement

Residuals must be redefined:

21/12-2000 (MJ)

Danish GPS Center

25

Results
Data fusion
5 Satellites GPS
0.33 deg

GPS + Sun sensor

0.30 deg

Improvement
9%

GPS + Magnetometer
0.32 deg

Improvement
3%

GPS + Rate gyros

0.32 deg

Improvement
3%

GPS + All
0.29 deg

Improvement
12 %

3 Satellites GPS
0.48 deg

GPS + All
0.38 deg

Improvement
21 %

21/12-2000 (MJ)

Danish GPS Center

26

Results
State fusion
Situation
Raw attitude from receiver Raw attitude from receiver + Kalman Raw attitude from receiver + Kalman + Secondary sensors Phase data from receiver + Kalman Phase data from receiver + Kalman + Secondary sensors

RSS error
0.49 deg 0.42 deg 0.37 deg 0.33 deg 0.29 deg

MFLOPS
0 5.9 20.2 18.6 50.6

21/12-2000 (MJ)

Danish GPS Center

27

Shadowed GPS
GPS + All secondary Pitch (u) Roll (v) Yaw (w) RSS
0.13 0.17 0.19 0.29

All secondary alone

0.26 0.56 0.63 0.88

21/12-2000 (MJ)

Danish GPS Center

28

Conclusion
Sensor fusion techniques can be applied to GPS applications mainly on the data and feature level The gain in accuracy was however shown to be only moderate in this project Significant error sources include less than optimal sensors and mechanical alignment problems The accuracy achieved by secondary sensors alone was shown to be sufficient for many space missions
21/12-2000 (MJ) Danish GPS Center 29