Subband domain signal validation

• US 7,085,675 B2
• Filed: 02/06/2003
• Issued: 08/01/2006
• Est. Priority Date: 02/06/2002
• Status: Expired due to Term

• Current Assignee: University of Chicago
• Original Assignee: University of Chicago
• Inventors: Wegerich, Stephan W.
• Primary Examiner(s): Bui, Bryan
• Assistant Examiner(s): PRETLOW, DEMETRIUS R
• Application Number:
  US10/360,843
• Publication Number:
  US 20030220767A1
• Time in Patent Office: 1,272 Days
• Field of Search: 702/180, 702/181, 702/179, 702/183
• US Class Current: 702/181
• CPC Class Codes: G05B 23/024 Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks
• CPC Class Codes: G06F 11/3452 Performance evaluation by statistical analysis

Abstract

A system, method and program product for predictive condition monitoring of a monitored system by means of sub-band decomposition of a complex monitored signal. Features of the decomposed signal are extracted for modeling estimation and detection of deviations in the features. Deviations are indicative of impending monitored system fault.

Claims

1. A method for early detection of a deviation from a selected condition in a monitored system comprising the steps of:

measuring a signal from the system containing information about operation of the system;

applying a plurality of filters to the signal for decomposing said signal into a plurality of sub-band components;

determining at least one feature value for at least one of said sub-band components;

generating an empirically-based estimate of said at least one feature value; and

comparing said estimate and said determined at least one feature value to detect a deviation from the selected condition.

2. A method according to claim 1, wherein said feature value is a scalar statistical function of a window of samples of the at least one sub-band component.

3. A method according to claim 1, wherein said feature value is selected from the set of mean, variance, skewness, kurtosis, root mean square, median, percentiles, auto-regression (AR) coefficients, ARMA coefficients, and auto-correlation coefficients.

4. A method according to claim 1, wherein said generating step includes referencing a learned state matrix of exemplar states.

5. A method according to claim 4, wherein the estimate is a linear combination of at least some exemplar states in said learned state matrix.

6. A method according to claim 5, wherein the linear combination is determined according to a similarity operation between a set of actual feature values and the at least some exemplar states.

7. A method according to claim 1 wherein said comparing step comprises:

generating a residual from a difference between said estimate and said determined feature value; and

performing a sequential probability ratio test on said residual to determine a statistically significant change.

8. An apparatus for monitoring a system for early detection of faults, comprising:

signal filter decomposition means disposed to decompose into a plurality of sub-band components a received signal indicative of operation of the system;

feature extraction means for determining at least one feature value for at least one of said plurality of sub-band components;

estimation means for generating an empirically-based estimate of said at least one feature value; and

alerting means disposed to compare said estimate and said determined feature value and generate an alert in response to a deviation between said estimate and said determined feature.

9. An apparatus according to claim 8, wherein said feature extraction means computes said feature value as a scalar statistical function of a window of samples of the at least one of the plurality of sub-band components.

10. An apparatus according to claim 8, wherein said estimation means comprises a learned state matrix of exemplar states.

11. An apparatus according to claim 10, wherein said estimation means generates an estimate from a linear combination of at least some of the exemplar states in said learned state matrix.

12. An apparatus according to claim 11, wherein said estimation means employs a similarity operation to determine the linear combination.

13. An apparatus according to claim 8 wherein said alerting means generates a residual from a difference between said estimate and said determined feature value, and performs a sequential probability ratio test on said residual to determine a statistically significant change.

14. A method for detection of a deviation from a selected operating condition of a monitored system, comprising the steps of:

measuring a signal from the monitored system with said signal containing information about operation of the monitored system;

creating a greater number of signal values from the signal in the form of a plurality of sub-band components;

determining at least one feature value for at least one of said sub-band components which are characteristic of the greater number of signal values;

generating an empirically-based estimate of said at least one feature value; and

comparing said empirically-based estimate and said at least one feature value to detect a deviation from the selected operating condition of the monitored system.

15. The method as defined in claim 14 wherein said feature value is a scalar statistical function of a window of samples of at least one sub-band component, the sub-band component being a function of the signal.

16. The method according to claim 14 wherein the signal comprises at least one sensed real data value.

17. The method as defined in claim 14 further including the step of applying a filter to the signal to create the plurality of sub-band components.

18. The method according to claim 14, wherein said generating step includes referencing a learned state matrix of exemplar states.

19. The method according to claim 18, wherein the estimate includes a combination of at least one of the exemplar states and the linear combination is determined according to a similarity operation between a set of actual feature values and the at least one of the exemplar states.

20. The method according to claim 14 wherein said comparing step comprises:

generating a residual from a difference between said estimate and said determined feature value; and

performing a sequential probability ratio test on said residual to determine a statistically significant change.