Monitoring method using kernel regression modeling with pattern sequences

• US 8,620,853 B2
• Filed: 07/19/2011
• Issued: 12/31/2013
• Est. Priority Date: 07/19/2011
• Status: Active Grant

• Current Assignee: SmartSignal Corporation (Emerson Electric Company)
• Original Assignee: SmartSignal Corporation (Emerson Electric Company)
• Inventors: Herzog, James P.
• Primary Examiner(s): Gaffin, Jeffrey A
• Assistant Examiner(s): Misir, Dave
• Application Number:
  US13/186,186
• Publication Number:
  US 20130024415A1
• Time in Patent Office: 896 Days
• Field of Search: None
• US Class Current: 706/52
• 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 18/22 Matching criteria, e.g. proximity measures

Abstract

A method for monitoring the condition of an object includes obtaining reference data that indicates the normal operational state of the object, and obtaining input pattern arrays. Each input pattern array has a plurality of time-ordered input vectors, while each input vector has input values representing a plurality of parameters indicating the current condition of the object. Then at least one processor generates estimate values based on a calculation that uses an input pattern array and the reference data to determine a similarity measure between the input values and reference data. The estimate values are compared to the corresponding input values so that resulting values from the comparison can be used to determine the condition of the object.

Claims

1. A method for monitoring the condition of an object, comprising:

obtaining reference data that indicates the normal operational state of the object;

obtaining input multi-dimensional pattern arrays, each input pattern array having a plurality of time-ordered input vectors, each input vector having input values representing a plurality of parameters indicating the current condition of the object;

generating, by at least one processor, estimate values based on a calculation that uses an input pattern array and the reference data to determine a similarity measure between the input values and reference data, the reference data being grouped in equal-sized and multi-dimensional training arrays, each of the training arrays being equal in size to a corresponding input multi-dimensional pattern array and each training array including a plurality of time-ordered reference vectors sequenced in time; and

comparing the estimate values to the corresponding input values so that resulting values from the comparison can be used to determine the condition of the object.

2. The method of claim 1, wherein at least one of the input vectors is in multiple input pattern arrays.

3. The method of claim 1, wherein each input pattern array represents a time period including a plurality of time points along a sequence of the input vectors representing the operational state of the object.

4. The method of claim 3, wherein each input pattern array represents a time period along the sequence that is different than, but overlaps, the time period of at least one of the other input pattern arrays.

5. The method of claim 3, wherein each input pattern array has a current primary vector from the sequence of input vectors and representing the most recent time point within the input pattern array.

6. The method of claim 5 comprising selecting input vectors for each input pattern array by moving a look-back window along the sequence, selecting a current primary vector at an end of the window, and selecting one or more older vectors that follow the primary vector along the sequence and fit within the window.

7. The method according to claim 6 wherein the older vectors are consecutive along the sequence.

8. The method according to claim 6 wherein the older vectors are not consecutive along the sequence.

9. The method according to claim 1 wherein time intervals between the time points represented by the input vectors within an input pattern array are uniform.

10. The method according to claim 1 wherein time intervals between the time points represent by the input vectors within an input pattern array are unequal.

11. The method according to claim 1 wherein time intervals between time points represented by the input vectors within an input pattern array are shorter between input vectors that are more recent than the time intervals between later input vectors within the same input pattern array.

12. The method according to claim 1 wherein the duration represented by the input pattern arrays varies.

13. The method according to claim 1 wherein the calculation is based on an equation that comprises:

14. The method according to claim 1 wherein the calculation is based on an equation that comprises:

15. The method of claim 1 wherein the similarity measure is a value on a bounded range that indicates an input value and reference value are identical at one end of the range and have no similarity at another end of the range.

16. The method of claim 1 comprising forming the reference data into at least one learned sequential pattern matrix, each learned sequential pattern matrix having a plurality of reference vectors, each reference vector having reference values representing a plurality of parameters.

17. The method of claim 16 wherein forming the reference data comprises forming a three-dimensional collection of the learned sequential pattern matrices.

18. The method of claim 17, wherein the three-dimensional collection of learned sequential pattern matrices is defined by three dimensions representing the parameters along a first dimension, the learned sequential pattern matrices along a second dimension, and time represented by the reference vectors in a third dimension.

19. The method of claim 17 wherein each input pattern array is compared to each learned sequential pattern matrix in the three-dimensional collection to determine a similarity measure for each learned sequential pattern matrix.

20. The method of claim 17, wherein each learned sequential pattern matrix represents a time period including time points represented by the reference vectors, wherein the time period is different than, but overlaps, the time period of at least some of the other learned sequential pattern matrices in the three-dimensional collection.

21. The method of claim 17 wherein the learned sequential pattern matrices have the same number of vectors as the input pattern arrays.

22. The method of claim 17 wherein the learned sequential pattern matrices represent the same time points as the input pattern arrays.

23. The method of claim 17 wherein the similarity between one of the learned sequential pattern matrices and one of the input pattern arrays comprises determining the similarity between corresponding reference and input values that represent the same vector position relative to the other vectors represented within their respective arrays.

24. The method of claim 17 wherein the similarity between one of the learned sequential pattern matrices and one of the input pattern arrays comprises calculating an average scalar similarity value for each parameter represented by the arrays.

25. The method of claim 17 wherein at least some of the input pattern arrays represent a time period that is different than the time period represented by at least one of the learned sequential pattern matrices.

26. The method according to claim 17 wherein the time period represented by the learned sequential pattern matrices varies.

27. The method according to claim 1 comprising generating the estimate values in the form of an estimate matrix having a plurality of estimate vectors.

28. The method of claim 27 wherein the estimate vectors in each estimate matrix corresponds to the time points represented by the input vectors in the input pattern arrays.

29. The method of claim 27 wherein less than all of the estimate values within the estimate matrix are used to determine the condition of the object.

30. The method of claim 27 comprising using only the estimate values that represent the most recent time point within the estimate matrix to determine the condition of the object.

31. The method of claim 27 comprising providing a single estimate vector for each time point represented across multiple estimate matrices to determine the condition of the object.

32. The method of claim 31 wherein the single estimate vector is provided by using at least one of;

selecting the estimate vector with one or more estimate values that has a maximum, minimum, or median value among the estimate vectors at the time point across multiple estimate matrices,

calculating an average, a weighted average, or weighted norm of the estimate values at the time point across multiple estimate matrices, and

selecting an estimate vector with the greatest similarity to a corresponding input vector among the estimate vectors at the time point across multiple estimate matrices.

33. The method of claim 27 comprising providing a single estimate vector for each estimate matrix to determine the condition of the object by using at least one of:

selecting an estimate vector within each estimate matrix that has a maximum, minimum, or median estimate value among the estimate vectors in the estimate matrix,

calculating an average, weighted average, or weighted norm of the estimate values in the estimate matrix, and

selecting the estimate vector within an estimate matrix with the greatest similarity to a respective corresponding input vector.

34. The method of claim 1 comprising generating inferred estimate values comprising using the similarity measures as weights in a calculation with a three-dimensional collection of learned sequential pattern matrices of the reference data that represents parameters that are not represented by the input pattern arrays.

35. The method of claim 34 wherein the reference data used to generate the similarity measures do not represent all of the parameters represented by the three-dimensional collection.

36. The method of claim 34 wherein the three-dimensional collection represents only parameters that are not represented in the input pattern arrays.

37. The method of claim 34 wherein the three-dimensional collection represents both parameters that are not represented in the input pattern arrays and parameters that are represented in the input pattern arrays.

38. The method of claim 1 wherein obtaining reference data comprises redetermining which reference data is to be used to generate the estimate values for each input pattern array.

39. The method of claim 1 comprising adapting an empirical model by adding the input pattern arrays to the reference data when the estimate values indicate the corresponding input pattern array represents normal operation.

40. A method for monitoring the condition of an object, comprising:

obtaining reference data that indicates the normal operational state of the object and in the form of a plurality of learned, equal-size, time sequential and multi-dimensional pattern matrices, each learned, equal-size, time sequential and multi-dimensional pattern matrix having a plurality of reference vectors, each reference vector having reference values representing a plurality of parameters, each of the plurality of learned, equal-size, time sequential and multi-dimensional pattern matrices being equal in size to a corresponding input multi-dimensional pattern array;

obtaining input data representing a plurality of parameters indicating the current condition of the object, the input data being in the form of a plurality of input multi-dimensional pattern arrays;

generating, by at least one processor, estimate values based on a calculation that uses the input data and the learned equal-size, time sequential and multi-dimensional pattern matrices to determine a similarity measure between the input data and reference values in the plurality of learned, equal-size, time sequential, and multi-dimensional pattern matrices;

comparing the estimate values to corresponding input values so that resulting values from the comparison can be used to determine the condition of the object.

41. A method for monitoring the condition of an object, comprising:

obtaining reference data that indicates the normal operational state of the object, the reference data being grouped in equal-sized and multi-dimensional training arrays, each of the training arrays being equal in size to a corresponding input multi-dimensional pattern array and each training array including a plurality of time-ordered reference vectors sequenced in time;

obtaining input data representing a plurality of parameters indicating the current condition of the object, the input data being a plurality of input multi-dimensional pattern arrays;

generating, by at least one processor, estimate values based on a calculation that uses both the input data and the reference data to determine similarity measures between the input data and the reference data, wherein the estimate values are generated in the form of an estimate matrix having a plurality of time-ordered estimate vectors, each estimate vector having estimate values representing multiple parameters; and

comparing at least one estimate vector for each time period represented by the estimate matrix to the input data so that the resulting values from the comparison can be used to determine the condition of the object.