Model-based fault detection and isolation for intermittently active faults with application to motion-based thruster fault detection and isolation for spacecraft

• US 20050143873A1
• Filed: 05/06/2004
• Published: 06/30/2005
• Est. Priority Date: 05/06/2003
• Status: Active Grant

• Current Assignee: Edward Wilson
• Original Assignee: Edward Wilson
• Inventors: Wilson, Edward
• Granted Patent: US 7,451,021 B2
• Time in Patent Office: Days
• US Class Current: 701/8
• CPC Class Codes: G05B 23/0227 Qualitative history assessment, whereby the type of data acted upon, e.g. waveforms, images or patterns, is not relevant, e.g. rule based assessment; if-then decisions
• CPC Class Codes: G05B 23/0254 based on a quantitative model, e.g. mathematical relationships between inputs and outputs; functions: observer, Kalman filter, residual calculation, Neural Networks

Abstract

The present invention is a method for detecting and isolating fault modes in a system having a model describing its behavior and regularly sampled measurements. The models are used to calculate past and present deviations from measurements that would result with no faults present, as well as with one or more potential fault modes present. Algorithms that calculate and store these deviations, along with memory of when said faults, if present, would have an effect on the said actual measurements, are used to detect when a fault is present. Related algorithms are used to exonerate false fault modes and finally to isolate the true fault mode. This invention is presented with application to detection and isolation of thruster faults for a thruster-controlled spacecraft. As a supporting aspect of the invention, a novel, effective, and efficient filtering method for estimating the derivative of a noisy signal is presented.

Claims

1. A method for detecting and isolating fault modes in a system having a model describing its behavior and one or more measurements that are sampled regularly, the method comprising:

a) said models and computing capacity to calculate past and present said measurements that would result from said system with no faults, as well as from said system with one or more potential said fault modes;

b) algorithms to calculate and store the deviations between the actual measurements and said calculated measurements;

c) memory of when said faults, if present, would have an effect on the said actual measurements;

d) detection algorithms using said calculated deviations and said memory to declare when one of said fault modes becomes true;

e) exoneration algorithms using said calculated deviations and said memory to remove certain fault modes from consideration as a potential fault mode, thereby making the decision making in the final step simpler and more robust; and

f) isolation algorithms using said calculated deviations and said memory to declare which one of the remaining said potential fault modes is the fault present.

2. The method of claim 1, wherein the system is controlled with on-off actuators and the fault modes are hard-off or hard-on actuator faults.

3. The method of claim 1, wherein the system is a thruster-controlled spacecraft, and the fault modes are thruster faults.

4. The method of claim 1, further including the case where direct comparison of measurements is replaced with comparison with an abstraction of the measurements such as returned by a filtering function.

5. The method of claim 1, further including the case where the model is replaced by a model adapted to match system outputs during periods where no failures are present, as with a neural network.

6. The method of claim 1, further including the case where instead of system fault modes, anomalies in the data un-related to system faults are detected.

7. The method of claim 1, further including the case where it is implemented sequentially, whereby once a fault mode is correctly detected and isolated, said method is re-initialized with this information to enable detection and isolation of any subsequently occurring faults.

8. The method of claim 1, wherein the application is to a system where a real-time controller exists, to be implemented either on the real-time processor, on a secondary processor as part of the system, or on a processor that is separate from the system, or by manual implementation of said method, where communication of required signals and results would occur for the latter two instances.

9. The method of claim 1, further including the case where excitation logic is used to adjust inputs to the system to improve the ability of said method to discern between a plurality of candidate fault modes, with said adjustments made as to minimize any negative impact on overall system performance.

10. The method of claim 3, wherein the method is implemented on-board the spacecraft using the main spacecraft processor.

11. The method of claim 3, wherein the method is implemented on-board the spacecraft using a secondary processor communicating with the main processor.

12. The method of claim 3, wherein the method is implemented off-board the spacecraft, including at a ground station, performing the said calculations based on telemetry from the spacecraft, and communicating the results back to the spacecraft with telemetry.

13. The method of claim 3, wherein said measurements are obtained from one or more of the following sensors:

gyroscopes of all varieties, accelerometers, star trackers, sun sensors, horizon sensors, video cameras, directional antennae, radar, and other measurements that directly or indirectly relate to spacecraft motions.

14. A method for efficiently and accurately estimating the average derivative of a noisy signal for which data is available for a plurality of samples during a sample set of interest, the method comprising:

a. a linear low-pass filter with zero-phase loss, used to attenuate high-frequency noise;

b. a least-squares line fit to the filtered data, used to determine the average derivative as the slope of the line, and the average value as the value at the midpoint of the line;

c. implementation of the above two steps as a vector multiplication of the said plurality of samples, thereby significantly reducing computational complexity and producing the same answer, as is possible since both steps are linear operations; and

d. a perturbation approach to automatically calculate the coefficients of said vector, thereby providing a fully automated means for updating the filter algorithm.

15. The method of claim 14, wherein information from outside said sample set of interest is also used to improve said derivative estimate, such as the line-fit of the preceding or following line segments.

16. The method of claim 14, wherein the function fit to the data is other than a line, such as a parabola, sinusoid, or superposition of a plurality of said functions.

17. The method of claim 14, further including the case where the objective of filtering is for operations other than derivative estimation;

parts (c) and (d) of the claim are applicable for implementation of a functional block consisting of an arbitrary combination of linear operations on a segment of data.