Trending system

• US 20050165520A1
• Filed: 02/22/2005
• Published: 07/28/2005
• Est. Priority Date: 01/28/2004
• Status: Active Grant

• Current Assignee: Honeywell International Inc.
• Original Assignee: Honeywell International Inc.
• Inventors: Ariyur, Kartik B.
• Granted Patent: US 7,474,992 B2
• Time in Patent Office: Days
• US Class Current: 701/29
• CPC Class Codes: G05B 23/0232 based on qualitative trend analysis, e.g. system evolution

Abstract

A trending system and method for trending data in a physical or clock system. The trending system includes a sliding window filter. The sliding window filter receives a data set of data points generated by the clock system. The sliding window filter partitions the data set into a plurality of data windows, and uses the data windows to calculate upper and lower confidence bounds for the data set. Specifically, the sliding window filter calculates upper confidence bounds and lower confidence bounds for each data point using each of the multiple data windows that includes the data point. The sliding window filter then selects the upper confidence bounds and the lower confidence bounds that result in the smallest mean prediction confidence interval for that data point. This results in a smoothed estimated trend for the data set that can be used for prognostication and fault detection.

Claims

1. A trending system comprising:

a sliding window filter; and

wherein;

the sliding window filter receives a data set from a clock system;

the data set comprises a plurality of data points;

the sliding window filter selects multiple data windows in the data set;

each of the data windows includes a subset plurality of the data points in the data set;

the sliding window filter generates estimates, upper confidence bounds and lower confidence bounds for each data point through regression and interval estimation over each of the multiple data windows that includes the data point;

the sliding window filter compares the interval widths of the confidence intervals produced by regression over each of the multiple data windows that includes the data point; and

the sliding window filter selects a fit for each data point that results in the smallest confidence interval for that data point.

2. The system of claim 1, wherein the sliding window filter outputs a filtered estimate of the data set.

3. The system of claim 1, further comprising:

a trend change detection mechanism; and

wherein;

the trend change detection mechanism determines a first convex hull for a set of upper confidence bounds and a second convex hull for a set of lower confidence bounds;

a comparator mechanism compares the first convex hull and the second convex hull to determine a transition point in the data set; and

an estimated trend of the data set is determined from the transition point and data after the transition point.

4. The system of claim 3, wherein the trend change detection mechanism compares the first convex hull to the second convex hull by determining if the first convex hull and the second convex hull intersect.

5. The system of claim 3, wherein the trend change detection mechanism compares the first convex hull and the second convex hull to determine a transition point in the data stream by determining if the first convex hull and the second convex hull intersect, and by iteratively discarding past points in the data set and generating a new first convex hull and a new second convex hull until there is no intersection between the new first convex hull and the new second convex hull.

6. The system of claim 1, further comprising:

an outlier elimination mechanism; and

wherein;

the outlier elimination mechanism removes statistical outliers in the data set by generating a first prediction cone for data points in a left sample window, generating a second prediction cone for data points in a right sample window, and determining if data points in a test window reside in the first prediction cone or the second prediction cone; and

a data point in the test window that does not reside in the first prediction cone or the second prediction cone is an outlier.

7. The system of claim 6, wherein the outlier elimination mechanism generates the first prediction cone and the second prediction cone by linear regression of the data points in the left sample window and linear regression of the data points in the right sample window.

8. The system of claim 7, wherein the outlier elimination mechanism moves the left sample window, right sample window, and test window through the data set to remove outliers through out the data set.

9. The system of claim 6, wherein the outlier elimination mechanism determines outliers by a weighted reciprocal of confidence intervals generated by prediction cones.

10. The system of claim 1, wherein adjacent windows in the multiple data windows overlap in the data set.

11. A method of trending data from a system, comprising:

receiving from the system a data set comprising a plurality of data points;

selecting multiple data windows in the data set, each of the data windows having a subset plurality of data points;

generating upper confidence bounds and lower confidence bounds for each of the data points using each of the multiple data windows that includes the data point; and

selecting an upper confidence bound and a lower confidence bound for each data point that results in the smallest confidence interval.

12. The method of claim 11, further comprising generating a filtered estimate of the data set from the selected upper confidence bounds and lower confidence bounds for each data point.

13. The method of claim 11, wherein the generating upper confidence bounds and lower confidence bounds comprises generating through a linear regression and a statistical inference.

14. The method of claim 13, wherein the statistical inference comprises using Student-t statistics.

15. The method of claim 11, further comprising:

generating a first convex hull from a first set of upper confidence bounds;

generating a second convex hull from a second set of lower confidence bounds; and

determining a transition point in the data set from a comparison of the first convex hull and the second convex hull.

16. The method of claim 15, wherein the determining a transition point in the data stream from the first convex hull and the second convex hull comprises determining if the first convex hull and the second convex hull intersect.

17. The method of claim 15, wherein the determining a transition point in the data stream from the first convex hull and the second convex hull comprises:

comparing the first convex hull to the second convex hull to determine if the first convex hull and the second convex hull intersect; and

iteratively discarding points in the data set and generating a new first convex hull and a new second convex full until there is no intersection between the new first convex hull and the new second convex hull.

18. The method of claim 11, further comprising removing outlier data from the data stream by generating a first prediction cone for data points in a left sample window, generating a second prediction cone for data points in a right sample window, and determining if data points in a test window reside in the first prediction cone or the second prediction cone, respectively.

19. The method of claim 18, wherein the first prediction cone and the second prediction cone are generated by linear regression of the data points in the left sample window and linear regression of the data points in the right sample window.

20. The method of claim 18, comprising moving the left sample window, right sample window and test window through the data set to remove outliers through out the data set.

21. The method of claim 11, wherein the multiple data windows comprise a left sample window, a test window and a right sample window that overlap in the data set.

22. The method of claim 11, wherein adjacent windows in the multiple data windows overlap in the data set.

23. A trending system comprising:

a trending program; and

signal bearing media bearing the trending program; and

wherein;

the trending program comprises a sliding window filter;

the sliding window filter receives a data set from a physical system;

the data set comprises a plurality of data points;

the sliding window filter selects multiple data windows in the data set;

each of the data windows has a subset plurality of the data points in the data set;

the sliding window filter generates upper confidence bounds and lower confidence bounds for each data point using each of the multiple data windows that includes the data point; and

the sliding window filter selects an upper confidence bound and a lower confidence bound for each data point that results in the smallest confidence interval for that data point.

24. The trending system of claim 23, wherein the signal bearing media comprises recordable media.

25. The trending system of claim 23, wherein the signal bearing media comprises transmission media.

26. The trending system of claim 23, wherein the sliding window filter outputs a filtered estimate of the data set.

27. The trending system of claim 23, wherein the sliding window filter determines the upper confidence bounds and lower confidence bounds through linear regression and statistical inference of the data set.

28. The trending system of claim 27, wherein the sliding window filter performs the statistical inference using Student-t statistics.

29. The trending system of claim 23, wherein:

the trending program further comprises a trend change detection mechanism;

the trend change detection mechanism determines a first convex hull for a set of upper confidence bounds and a second convex hull for a set of lower confidence bounds;

a comparator mechanism compares the first convex hull and the second convex hull to determining a transition point in the data set; and

an estimated trend of the data set is determined from the transition point.

30. The trending system of claim 29, wherein the trend change detection mechanism compares the first convex hull to the second convex hull by determining if the first convex hull and the second convex hull intersect.

31. The trending system of claim 29, wherein the trend change detection mechanism compares the first convex hull and the second convex hull to determine a transition point in the data stream by determining if the first convex hull and the second convex hull intersect, and by iteratively discarding points in the data set and generating a new first convex hull and a new second convex hull until there is no intersection between the new first convex hull and the new second convex hull.

32. The trending system of claim 23, wherein:

the trending program further comprises an outlier elimination mechanism;

the outlier elimination mechanism removes statistical outliers in the data set by generating a first prediction cone for data points in a left sample window, generating a second prediction cone for data points in a right sample window, and determining if data points in a test window reside in the first prediction cone or the second prediction cone; and

a data point in the test window that does not reside in the first prediction cone or the second prediction cone is an outlier.

33. The trending system of claim 32, wherein the outlier elimination mechanism generates the first prediction cone and the second prediction cone by linear regression of the data points in the left sample window and linear regression of the data points in the right sample window.

34. The trending system of claim 33, wherein the outlier elimination mechanism moves the left sample window, right sample window, and test window through the data set to remove outliers through out the data set.

35. The trending system of claim 32, wherein the outlier elimination mechanism determines outliers by a weighted reciprocal of confidence intervals generated by prediction cones.

36. The trending system of claim 23, wherein adjacent windows in the multiple data windows overlap in the data set.