Methods for condition monitoring and system-level diagnosis of electro-mechanical systems with multiple actuating components operating in multiple regimes

• US 6,539,343 B2
• Filed: 02/03/2000
• Issued: 03/25/2003
• Est. Priority Date: 02/03/2000
• Status: Expired due to Term

• Current Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Original Assignee: Xerox Corporation (Xerox Holdings Corp.)
• Inventors: Zhao, Feng, Manders, Eric-J.
• Primary Examiner(s): ASSOUAD, PATRICK J
• Application Number:
  US09/497,113
• Publication Number:
  US 20020103626A1
• Time in Patent Office: 1,146 Days
• Field of Search: 702/189, 702/190, 708/400, 706/911, 700/30, 700/47, 700/51, 703/6, 703/13, 703/22
• US Class Current: 702/190
• 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
• CPC Class Codes: G06F 17/148 Wavelet transforms

Abstract

The present invention discloses methods for condition-based monitoring of a system and its components using joint time-frequency analysis and signal demodulation. A first method allows construction of models of non-stationary signals having multiple time and frequency scales of interest. In particular, the method uses wavelet analysis to encode essential information in time-frequency transformations and provides a signal model for an operating fault or normal operation of a system of interest. A second method uses the signal models for condition-based monitoring to detect a change in a signal. In this method, the signal models are correlated with sensor signals to determine the operating condition of a system or system component. A third method for condition-based monitoring uses signal demodulation. In this method, a nominal response of a system component is subtracted from a detected signal to more reliably detect subtle changes in the signal due to component degradation.

Claims

1. A method for constructing time-frequency models of non-stationary signals comprising the steps:

collecting a plurality of training data from at least one sensor embedded in a system of interest, wherein the training data includes an operating fault of the system;

extracting wavelet coefficient features from the training data by performing wavelet analysis;

constructing a signal template of the operating fault using the wavelet coefficients.

2. The method of claim 1, wherein the signal template comprises a sequence of wavelet coefficients that represent an onset of the operating fault signal, a duration of the operating fault signal, and a magnitude of wavelet components described by the sequences of wavelet coefficients.

3. The method of claim 1, wherein the training data includes a plurality of known operating faults.

4. The method of claim 1 wherein the training data includes nominal operation of the system.

5. The method of claim 1, wherein the plurality of training data is repeatedly collected over a fixed window of time.

6. The method of claim 1, further including the step of collecting training data from a plurality of sensors responsive to a plurality of stimuli.

7. The method of claim 1, further including the step of computing wavelet coefficient features from multiple training data segments corresponding to an operating condition and statistically averaging the features to form a signal template for the operating condition.

8. The method of claim 1, further including the step of constructing a signal template for the normal operation of the system from the training data.

9. The method of claim 1, further including the step of constructing a plurality of signal templates for the plurality of known operating faults from the training data.

10. The method of claim 6, further including the step of building a library of signal templates of operating faults of the system of interest from the plurality of sensor signals.

11. A method for condition-based monitoring of a system of interest comprising the steps:

providing signal templates representing a normal operating condition and a faulty operating condition of a system of interest, wherein the signal templates comprise wavelet coefficients;

constructing sensor signal segments by extracting wavelet coefficient features using wavelet analysis from a signal from a sensor monitoring the system of interest, wherein the sensor is embedded in the system of interest;

convolving the signal template with the segments of the signal from the sensor monitoring the system of interest; and

classifying the segments of the sensor signal into the normal and faulty operating conditions of the system based on correlation amplitude.

12. The method of claim 11, wherein the signal from the sensor is collected from a moving window of fixed time duration.

13. The method of claim 11, wherein wavelet coefficient features are extracted using wavelet analysis from a plurality of signals from a plurality of sensors monitoring the system of interest.

14. The method of claim 13, wherein the signal templates are correlated with a plurality of signals from a plurality of sensors.

15. The method of claim 14, wherein the signal templates represent normal and faulty operating conditions for a plurality of components of the system of interest.

16. The method of claim 11, wherein a discriminant analysis method is used to classify the sensor signals.

17. The method of claim 11, wherein the signal templates comprise wavelet coefficients constructed from training data.

18. The method of claim 11, wherein the step of classification of sensor signals uses decision procedures.

19. A method for condition-based monitoring of a component of a system comprising the steps:

constructing a signal model of the nominal response of the component from training data;

detecting a signal from a sensor monitoring the component; and

subtracting the signal model from the signal from the sensor monitoring the system to create a residual signal.

20. The method of claim 19, wherein the signal model for the nominal response is determined by statistically averaging multiple trained data segments corresponding to the normal operating condition.

21. The method of claim 20, wherein the signal model for the nominal responses is a signal template of normal operation of the system constructed from training data.

22. The method of claim 19, wherein the signal model for the nominal response is determined by a statistical regression on training data corresponding to the normal operating condition.

23. The method of claim 19, wherein the signal model for the nominal response is determined by a lumped parameter model of the component under normal operating condition, wherein the model parameters are estimated by a statistical regression.

24. The method of claim 19, further including the step of analyzing the residual signal to detect changes in the sensor signal.

25. The method of claim 24, wherein the residual signal is analyzed by classifying the sensor signal using signal templates of normal and faulty operating conditions.

26. A method for condition-based monitoring of a system having a plurality of components, comprising the steps of:

constructing a signal model of the nominal response for each of plurality of components in the system;

detecting signals from a plurality of sensors monitoring the plurality of components; and

subtracting the signal models from each signal from the plurality of sensors monitoring the components to create a residual signal for the sensor.

27. The method of claim 26, wherein the nominal responses of the components result from excitation from a plurality of actuators.

28. A method for condition-based monitoring of a system having a plurality of components, comprising the steps of:

constructing a signal model of the nominal response at each of a plurality of sensors monitoring a plurality of components;

detecting signals from a plurality of sensors monitoring the plurality of components; and

subtracting the signal models from the signal from the corresponding sensor monitoring the components.