Download as pdf or txt
Download as pdf or txt
You are on page 1of 21

INSTITUTE FOR SYSTEMS AND ROBOTICS

Fault Detection and Isolation for Inertial Measurement Units


S. Brs P. Rosa C. Silvestre P. Oliveira

Institute for Systems and Robotics Instituto Superior Tcnico Lisbon, Portugal

51st IEEE Conference on Decision and Control Maui, Hawaii, USA

Introduction
The navigation system is a critical component in any aircraft or spacecraft.

In high reliability systems it is not only necessary to detect faults, but also to isolate the defective sensor.

Objective:
Development of Fault Detection and Isolation (FDI) strategies for Inertial Measurement Units (IMUs).

Introduction
Two strategies are proposed, which exploit:
Hardware redundancy by identifying non-compatible redundant sensor measurements. Analytical redundancy by using Set-Valued Observers to model the dynamic relation between measurement of the sensor suite. It is assumed that the sensor measurements are corrupted by bounded noise. The proposed solutions guarantee that there will be no false alarms. Tuning of a decision rule based on a threshold to declare whether or not a fault has occurred is not needed.

Sensor Model
Rate-gyros model

Inertial vector model (such as accelerometers)

Each component of

satisfies

Hardware Redundancy
Measurement model Matrix form

Hardware Redundancy

(k)
Non-faulty sensors

(k)
Faulty sensor

Hardware Redundancy
Fault detection and isolation algorithm:
If Set(MH,m) is empty, a fault is detected. Then the emptiness of Si=Set(MH\{i},m\{i}) for all sensors is evaluated. If only for one i, Si is non-empty, the faulty measurement is r(i). If not, it is not possible to isolate the fault at this instant.

Hardware Redundancy
Proposition:
If the magnitude of the fault is greater than , the proposed FDI scheme is able to detect and isolate non-simultaneous faults.

Assumption:
There are at least five sensor measurements.

Set-Valued Observers

There is uncertainty on the state and measurements. Uncertainty is described by means of polytopes. The dynamic model may be uncertain.

on

the

Set-Valued Observers

Prediction (model)

Update
(measurements)

10

Analytical Redundancy
Analytical model of the vector observations:

Solution:

Using the Mean Value Theorem it can be concluded that

Bounds on the transition matrix elements

11

Analytical Redundancy
Fault detection filter:

12

Analytical Redundancy
Fault detection and isolation filter:

13

Analytical Redundancy
Stages of the FDI filter:

14

Simulation Results
Angular velocity:

Vector measurements:

Noise bounded by 0.05 Sampling period of all sensors set to T=0.1 s 100 Monte-Carlo runs

15

Simulation Results
We assume that one of the following six faults can occur:

1. a stuck at type of fault in rate gyro #1; 2. rate gyro #3 badly damaged generating a null measurement; 3. the maximum amplitude of the noise in the rate gyro #3 increases fifteen times; 4. the second component of vector #2 is null; 5. a stuck at type of fault in first sensor of vector #1; 6. the maximum amplitude of the noise in the third sensor of vector #3 increases five times.

16

Simulation Results
Five gyros and five sensors per vector

Three gyros and three sensors per vector

17

Simulation Results

18

Simulation Results

19

Conclusions
Two novel FDI methodologies for IMUs and vector observations were proposed. The first scheme takes advantage of hardware redundancy in the sensor measurements to detect incoherences between them. Sufficient conditions have been provided that guarantees isolation of nonsimultaneous faults. The dynamic relation between the angular velocity and the vector measurements is exploited in the second methodology, which resorts to the setvalued state estimates provided by SVOs to validate or falsify different models of faults. Neither solution generates false detections, as long as the non-faulty model of the system remains valid. Simulation results show that the detection and isolation of the faults take, in general, only a few iterations.
20

Thank you

21

You might also like