Monitoring and fault detection system and method using improved empirical model for range extrema

• US 20020152056A1
• Filed: 02/22/2001
• Published: 10/17/2002
• Est. Priority Date: 02/22/2001
• Status: Active Grant

• Current Assignee: SmartSignal Corporation (Emerson Electric Company)
• Original Assignee: SmartSignal Corporation (Emerson Electric Company)
• Inventors: Wegerich, Stephan W., Herzog, James P.
• Granted Patent: US 7,373,283 B2
• Time in Patent Office: Days
• US Class Current: 703/2
• CPC Class Codes: G06F 17/18 for evaluating statistical data , e.g. average values, frequency distributions, probability functions, regression analysis forecasting specially adapted for a specific administrative, business or logistic context G06Q10/04

Abstract

A method and apparatus for improved monitoring the operational state of instrumented systems is provided. An empirical model characterizes normal or desirable operation of the system, and real-time observations are provided to the model to generate estimates of expected sensor values. Comparison of the estimates with the real-time observations provides advanced warning of discrepancies in the operational state of the instrumented system. The invention provides for incipient failure detection, sensor failure detection and incipient process upset. An improved similarity operator provides for estimates that are not impaired by real-time observations at or beyond the limits of modeled data. The similarity operator comprises a difference function added to a constant, and the result is inverted.

Claims

1. A method for calculating a similarity value, comprising:

providing a first state vector;

providing a second state vector, the first state vector and the second state vector having at least one pair of corresponding elements;

calculating a function value for at least one pair of corresponding elements of said first and second state vectors; and

calculating a similarity value for the at least one pair of corresponding elements, having a denominator comprising said function value.

2. The method of claim 1, wherein the similarity value is calculated as an inverse of the sum of the function value and a predetermined value.

3. The method of claim 2, wherein the predetermined value is one.

4. The method of claim 1, wherein the function value is raised to a power and divided by a constant inside the denominator.

5. The method of claim 4, wherein said power is selected in the range of one to four.

6. The method of claim 4, wherein said dividing constant is in the range of one to four.

7. The method of claim 2, wherein the function value equals zero when a pair of corresponding elements are identical.

8. The method of claim 7, wherein the function value increases with increasing difference of a pair of corresponding elements.

9. The method of claim 8, wherein the function value is calculated by dividing the absolute difference of a pair of corresponding elements by an expected range.

10. The method of claim 8, wherein the function value is calculated by subtracting from a first angle defining a similarity domain a second angle defined by the relative locations with respect to the similarity domain of a pair of corresponding elements.

11. The method of claim 1, wherein a plurality of similarity values, each of the plurality of similarity values corresponding to one a pair of corresponding elements, are combined to provide a vector-to-vector similarity.

12. The method of claim 11, wherein said plurality of similarity values are combined by averaging them.

14. The computable program product of claim 13, wherein the program code means for computing the elemental similarity computes it according to:

15. The computable program product of claim 14, wherein the program code means for computing the difference function computes it according to:

16. The computable program product of claim 13, further comprising program code means for generating an estimated state vector from a composite of reference library state vectors using vector similarities of each state vector to the monitored state vector.

17. The computable program product of claim 13, further comprising program code means for determining a highest vector similarity among vector similarities of each state vector in the reference library to the monitored state vector, and selecting for output a classification associated with the reference state vector having the highest vector similarity.

19. The system of claim 18, wherein the current input matrix includes a plurality of values, the reference matrix includes a plurality of values, and the comparison means further comprises:

means for locating corresponding values in the plurality of values in the reference matrix for at least one of the plurality of values in the current input matrix;

means for calculating an absolute difference between one of the plurality of values in the current input matrix with each of the corresponding values in the plurality of values in the reference matrix;

means for calculating an difference between a maximum value and a minimum value among the corresponding values in the plurality of values in the reference matrix.

20. The system of claim 18, wherein the function value is calculated by dividing the absolute difference by an expected range.

21. The system of claim 18, wherein the similarity value is calculated as an inverse of the sum of the function value and a predetermined value.

13. A computable program product for computing similarity values for monitoring and classifying states of operation of a system, comprising:

program code means for selecting a next state vector from a reference library of state vectors comprising elemental parameter values from a memory;

program code means for retrieving a monitored state vector comprising elemental parameter values from an input source;

program code means for computing a difference function for pairs of corresponding elements selected from the next state vector and the monitored state vector;

program code means for computing an elemental similarity for pairs of corresponding elements, having the difference function result in its denominator; and

program code means for combining elemental similarities for all pairs of corresponding elements to provide a vector similarity of the next state vector and the monitored state vector.

18. A system for detecting a difference in a signal, comprising:

means for monitoring a signal;

means for receiving a current input matrix, the current input matrix being based on the signal;

means for comparing the current input matrix to a reference matrix;

means for calculating a function value, the function value being based on the comparison of the current input matrix with the reference matrix;

means for calculating a similarity value, the similarity value having the function value in its denominator; and

means for generating a status signal based on the similarity value.

22. An apparatus for monitoring a signal, comprising:

a sensor for receiving a plurality of current input values;

a memory location for storing a plurality of reference values; and

a processor coupled to the sensor, the processor being disposed to receive the plurality of current input values and to calculate a similarity value based on a comparison between each of the plurality of current input values and each of the plurality of reference values, wherein the result of the comparison is inversely related to the similarity value.

23. The apparatus of claim 22, wherein the similarity value is calculated as a numerical closeness of each value in the current input matrix received from the sensor to each value in the reference matrix stored in the memory location.

24. The apparatus of claim 22, wherein the comparison is a function of the absolute difference between each of the plurality of current input values and each of the plurality of reference values divided by the difference between a maximum value of each of the plurality of the reference values and a minimum value of each of the plurality of the reference values.

26. The method of claim 25, further comprising the step of combining the similarities for corresponding parameters of the monitored snapshot and a reference library snapshot into a snapshot similarity.

27. The method of claim 26, further comprising the step of generating an estimate snapshot of said parameters based on at least one snapshot similarity.

28. The method of claim 27, further comprising the step of comparing the estimate snapshot to the monitored snapshot and generate an alert on determining a difference.

29. The method of claim 26, further comprising:

providing a classification associated with at least some of the reference library snapshots; and

determining a classification for said monitored snapshot based on said snapshot similarities and the reference library snapshot classifications.

30. The method of claim 29, wherein said determining step comprises determining the class of the reference library snapshot having the highest snapshot similarity with said monitored snapshot.

25. A method for empirically modeling multiple parameters of a system selected from a process, a machine and a biological system, comprising:

providing a reference library of snapshots of time-related values of said parameters representing operation of said system;

providing a monitored snapshot of values of at least some of said parameters;

computing for the at least some parameters values of a difference function of corresponding parameters of said monitored snapshot and said reference library; and

computing for the at least some parameters similarities of corresponding parameters as the inverse of a constant plus the value of the difference function.

31. An apparatus for monitoring the state of a system instrumented with sensors, selected from a process or a machine, comprising:

a memory for storing a reference library of snapshots of time-related values of said sensors representing operation of said system;

means for acquiring a monitored snapshot of sensor values from the system;

a similarity engine disposed to receive the monitored snapshot from said data acquisition means and generate similarity values for comparisons of the monitored snapshot to reference library snapshots in said memory; and

output means for indicating a system state based on the similarity values;

where each similarity value for a comparison of the monitored snapshot to a reference library snapshot is a statistical composite of similarity values for corresponding elements of the two snapshots, derived from the inverse of a quantity comprising a constant plus a difference function of the corresponding elements.

32. The apparatus of claim 31, further comprising an estimated state generator responsive to the generated similarity values for generating a snapshot of estimated sensor values.

33. The apparatus of claim 32, wherein the estimated state generator generates the estimated sensor values as a composite of the reference library snapshots, weighted by corresponding similarity values from said similarity engine.

34. The apparatus of claim 32, further comprising a deviation detection engine disposed to receive the estimated sensor values from said estimated state generator and the monitored snapshot, and generate an indication of a difference between them.

35. The apparatus of claim 34, wherein the deviation detection engine differences the estimated sensor values with the monitored snapshot to produce a residual snapshot, and performs a sequential probability ratio test on at least one sensor across a sequence of such residual snapshots.

36. The apparatus of claim 31, wherein said output means outputs a classification associated with at least one reference library snapshot, based on which reference library snapshot has the highest similarity value with the monitored snapshot.