Adaptive modeling of changed states in predictive condition monitoring

• US 20020133320A1
• Filed: 02/27/2001
• Published: 09/19/2002
• Est. Priority Date: 01/19/2001
• Status: Active Grant

• Current Assignee: SmartSignal Corporation (Emerson Electric Company)
• Original Assignee: SmartSignal Corporation (Emerson Electric Company)
• Inventors: Xu, Xiao, Wegerich, Stephan W., Bell, David R.
• Granted Patent: US 7,233,886 B2
• Time in Patent Office: Days
• US Class Current: 703/2
• 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: G05B 23/0297 Reconfiguration of monitoring system, e.g. use of virtual sensors; change monitoring method as a response to monitoring results

Abstract

An improved empirical model-based surveillance or control system for monitoring or controlling a process or machine provides adaptation of the empirical model in response to new operational states that are deemed normal or non-exceptional for the process or machine. An adaptation decision module differentiates process or sensor upset requiring alerts from new operational states not yet modeled. A retraining module updates the empirical model to incorporate the new states, and a pruning technique optionally maintains the empirical model by removing older states in favor of the added new states recognized by the model.

Claims

1. A method for adapting an empirical model used in monitoring a system, comprising the steps of:

receiving actual values of monitored parameters characterizing the system;

generating from the empirical model estimates of the parameter values in response to receiving actual values;

calculating a global similarity score for a comparison of a set of estimated parameter values and a related set of actual parameter values; and

adapting the empirical model to account for at least some received actual values based on the global similarity score.

2. A method according to claim 1 where said adapting step comprises adapting the empirical model when a global similarity score falls outside of a selected range.

3. A method according to claim 1 where in the adapting step comprises:

identifying a beginning of a transition phase when a global similarity score falls outside of a selected range;

identifying an end of the transition phase; and

determining in response to identification of the end of the transition phase whether the system is in a new state not accounted for in the empirical model; and

updating the empirical model to account for the new state in response to a determination that the new state was not accounted for in the empirical model.

4. A method according to claim 3 wherein the determining step comprises comparing at least one global similarity score after the end of the transition phase to a threshold and if it is below the threshold then concluding that the new state is not accounted for in the empirical model.

5. A method according to claim 3 wherein the determining step comprises comparing a window of global similarity scores after the end of the transition phase to a threshold and if at least a selected number of said scores in said window fall below the threshold then concluding that the new state is not accounted for in the empirical model.

6. A method according to claim 3 wherein the determining step comprises comparing the mean of a window of global similarity scores after the end of the transition phase to a threshold and if the mean of said scores in said window falls below the threshold then concluding that the new state is not accounted for in the empirical model.

7. A method according to claim 3 wherein the empirical model has a reference library of snapshots of parameter values characterizing recognized system states, and the determining step comprises comparing a snapshot of received actual values after the end of the transition phase to each snapshot in the reference library to compute a similarity for each comparison, and if the highest such similarity is less than a selected threshold then concluding that the new state is not accounted for in the empirical model.

8. A method according to claim 3 wherein the empirical model has a reference library of snapshots of parameter values characterizing recognized system states, and the determining step comprises comparing each in a series of snapshots of received actual values after the end of the transition phase to each snapshot in the reference library to compute a similarity for each comparison, and if the mean over the series of the highest similarity for each actual snapshot is less than a selected threshold then concluding that the new state is not accounted for in the empirical model.

9. A method according to claim 3 wherein the empirical model has a reference library of snapshots of parameter values characterizing recognized system states, and the determining step comprises comparing each in a series of snapshots of received actual values after the end of the transition phase to each snapshot in the reference library to compute a similarity for each comparison, and if the highest similarity of at least a selected number of actual snapshots over the series is less than a selected threshold then concluding that the new state is not accounted for in the empirical model.

10. A method according to claim 3, wherein the step of identifying the end of the transition phase comprises examining a moving window of successive global similarity scores and when each of at least a selected number of successive global similarity scores in the window lies within a selected range around the previous global similarity score, identifying that the end of transition has been reached.

11. A method according to claim 3, wherein the step of identifying the end of the transition phase comprises examining a moving window of successive global similarity scores and when each of at least a selected number of successive global similarity scores in the window lies within a selected range around the mean of the previous global similarity scores in the window, identifying that the end of transition has been reached.

12. A method according to claim 3 wherein the empirical model has a reference library of snapshots of parameter values characterizing recognized system states, and the updating step comprises adding at least one snapshot of the received actual values to the reference library.

13. A method according to claim 1 wherein the selected range has a lower threshold equal to a multiple of a standard deviation for a sequence of global similarity scores, subtracted from the mean of the sequence of global similarity scores.

14. A method according to claim 1 wherein the calculating step comprises comparing corresponding elements of a snapshot of received actual parameter values and a related snapshot of estimates to generate elemental similarities, and statistically combining the elemental similarities to generate the global similarity score.

16. A method according to claim 15 wherein the reference data set comprises snapshots of time-correlated parameter values, and said adapting step comprises adding at least one snapshot of the received actual values to the reference set.

18. An apparatus according to claim 17 where said adaptation module is disposed to adapt the empirical model when a global similarity score falls outside of a selected range.

19. An apparatus according to claim 18 where said empirical model comprises a reference library of snapshots of time-correlated parameter values, and said adaptation module adapts the empirical model by adding at least one snapshot of the received actual values to the reference library.

20. An apparatus according to claim 17 wherein the adaptation module first identifies a transition phase after which it adapts the empirical model to account for at least some received actual parameter values based on global similarity scores from said global similarity engine.

21. An apparatus according to claim 20 wherein the adaptation module identifies a beginning of the transition phase when a global similarity score falls outside of a selected range.

22. A method according to claim 20, wherein the adaptation module identifies the end of the transition phase by examining a moving window of successive global similarity scores and when each of at least a selected number of successive global similarity scores in the window lies within a selected range around the previous global similarity score, identifying that the end of transition has been reached.

23. A method according to claim 20, wherein the adaptation module identifies the end of the transition phase by examining a moving window of successive global similarity scores and when each of at least a selected number of successive global similarity scores in the window lies within a selected range around the mean of the previous global similarity scores in the window, identifying that the end of transition has been reached.

24. An apparatus according to claim 20 wherein the adaptation module determines in response to identification of the transition phase whether the system is in a new state not accounted for in the empirical model

25. An apparatus according to claim 24 wherein the adaptation module determines that the new state is not accounted for in the empirical model if at least one global similarity score after the end of the transition phase is below a selected threshold.

26. An apparatus according to claim 24 wherein the adaptation module determines that the new state is not accounted for in the empirical model if at least a selected number of global similarity scores in a window of global similarity scores after the end of the transition phase fall below a selected threshold.

27. An apparatus according to claim 24 wherein the adaptation module determines that the new state is not accounted for in the empirical model if the mean global similarity value over a window of global similarity scores after the end of the transition phase falls below a selected threshold.

28. An apparatus according to claim 24 wherein the empirical model comprises a reference library of snapshots of parameter values characterizing recognized system states, and the adaptation module compares a snapshot of received actual values after the end of the transition phase to each snapshot in the reference library to compute a similarity for each comparison, and if the highest such similarity is less than a selected threshold then determines that the new state is not accounted for in the empirical model.

29. An apparatus according to claim 24 wherein the empirical model comprises a reference library of snapshots of parameter values characterizing recognized system states, and the adaptation module compares each in a series of snapshots of received actual values after the end of the transition phase to each snapshot in the reference library to compute a similarity for each comparison, and if the mean over the series of the highest similarity for each actual snapshot is less than a selected threshold then determines that the new state is not accounted for in the empirical model.

30. An apparatus according to claim 24 wherein the empirical model comprises a reference library of snapshots of parameter values characterizing recognized system states, and the adaptation module compares each in a series of snapshots of received actual values after the end of the transition phase to each snapshot in the reference library to compute a similarity for each comparison, and if the highest similarity of at least a selected number of actual snapshots over the series is less than a selected threshold then determines that the new state is not accounted for in the empirical model.

31. An apparatus according to claim 21 wherein the selected range has a lower threshold equal to a multiple of a standard deviation for a sequence of global similarity scores, subtracted from the mean of the sequence of global similarity scores.

15. A method for adapting a reference data set of an empirical model used in empirically modeling a system, comprising the steps of:

receiving actual values of monitored parameters characterizing the system, including at least one control parameter;

determining that the actual value for the at least one control parameter lies outside a predetermined range;

adapting the empirical model in response to said determining step, to account for the received actual values.

17. An apparatus for monitoring the operation of a system characterized by operational parameters, comprising:

an empirical model for generating estimates of parameter values in response to receiving actual parameter values characterizing the operation of the system;

a global similarity engine for generating a global similarity score for a comparison of a set of estimated parameter values and a related set of actual parameter values from said system; and

an adaptation module for adapting the empirical model to account for at least some received actual parameter values based on the global similarity score.

32. A computer program product for monitoring the operation of a system characterized by operational parameters, comprising:

computer readable program code means for receiving actual parameter values characterizing operation of the system, including at least one control parameter;

computer readable program code means for determining that an actual value for the at least one control parameter lies outside a predetermined range;

data storage means for storing data representing an empirical model of the system; and

computer readable program code means for adapting the stored empirical model data in response to said determining step, to account for the received actual values.

33. A computer program product according to claim 32 wherein the empirical model data comprises snapshots of time-correlated parameter values, and said means for adapting is disposed to add at least one snapshot of the received actual parameter values to the empirical model data.

34. A computer program product according to claim 33 further comprising computer readable program means for generating estimates of parameter values in response to the means for receiving, using the empirical model data in said data storage means.

35. A computer program product according to claim 34 wherein said means for estimating uses a similarity operator.