System of sequential kernel regression modeling for forecasting and prognostics

• US 9,250,625 B2
• Filed: 07/19/2011
• Issued: 02/02/2016
• Est. Priority Date: 07/19/2011
• Status: Active Grant

• Current Assignee: GE Intelligent Platforms Incorporated (Emerson Electric Company)
• Original Assignee: GE Intelligent Platforms Incorporated (Emerson Electric Company)
• Inventors: Herzog, James P.
• Primary Examiner(s): Gaffin, Jeffrey A
• Assistant Examiner(s): Kim, David H
• Application Number:
  US13/186,153
• Publication Number:
  US 20130024414A1
• Time in Patent Office: 1,659 Days
• Field of Search: 706/52
• US Class Current: 1/1
• CPC Class Codes: G05B 23/0243 model based detection method, e.g. first-principles knowledge model
• CPC Class Codes: G06F 18/22 Matching criteria, e.g. proximity measures
• CPC Class Codes: G06N 7/023 Learning or tuning the parameters of a fuzzy system

Abstract

A monitoring system for determining the future operational condition of an object includes an empirical model to receive reference data that indicates the normal operational state of the object and input pattern arrays. Each input pattern array has a plurality of input vectors, while each input vector represents a time point and has input values representing a plurality of parameters indicating the current condition of the object. The model 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, in the form of an estimate matrix, include at least one estimate vector of inferred estimate values, and represents at least one time point that is not represented by the input vectors. The inferred estimate values are used to determine a future condition of the object.

Claims

1. A monitoring system for determining the future operational condition of an object, comprising:

an empirical model module configured to;

receive reference data that indicates the normal operational state of the object,

receive input multi-dimensional pattern arrays, each input pattern array having a plurality of input vectors, each input vector representing a time point and having input values representing a plurality of parameters indicating the current condition of the object obtained from one or more first sensors, each input vector arranged and maintained in a predetermined and time ordered relationship with the others, and

generate 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 similarity measure accounting for the predetermined and ordered time relationship wherein the estimate values are in the form of an estimate matrix that includes at least one estimate vector of inferred estimate values for at least one future point in time or a plurality of second sensors being different from the one or more first sensors, each estimate vector arranged and maintained in the predetermined and ordered time relationship with the others, the reference data being grouped in equal-sized and multi-dimensional training arrays including reference vectors, such that each reference vector in any training array is arranged and maintained in the predetermined and ordered time relationship with the others and such that the generation does not destroy any time information of the estimate matrix or the reference data; and

a prognostic module configured to use the inferred estimate values to determine a future condition of the object.

2. The system of claim 1 wherein the estimate matrices only include estimate vectors that represent time points that are not represented by the input vectors.

3. The system of claim 1 wherein the estimate matrices include at least one estimate vector that represents the same time point represented by the input vectors and at least one estimate vector that represents a time point that is not represented by the input vectors.

4. The system of claim 1 wherein the estimate matrices include estimate values that represent parameters that indicate the condition of the object and that are not represented by the input values.

5. The system of claim 1 wherein each estimate matrix represents a primary current time point and time points not represented by the input vectors and that are succeeding time points relative to the current time point.

6. The system of claim 1 wherein the empirical module is configured to generate weight values by using the similarity measures, and uses the weight values in a calculation with the reference data to generate the estimate matrix.

7. The system of claim 6 wherein the weights values are in the form of a weight vector.

8. The system of claim 6 wherein the reference data used in the calculation with weight values comprises reference values that represent time points that are not represented by the input pattern arrays.

9. The system of claim 8 wherein the reference data used in the calculation with weight values represents a primary current time point and the time points not represented by the input vectors are succeeding time points relative to the current time point.

10. The system of claim 6 wherein the reference data used in the calculation with the weight values is in the form of a three-dimensional collection of learned sequential pattern matrices, each learned sequential pattern matrix comprising reference vectors of reference values, wherein each reference vector represents a different time point within the learned sequential pattern matrix.

11. The system of claim 10 wherein each learned sequential pattern matrix comprises a primary current time point and time points that represent succeeding time points relative to the primary current time point and that are not represented by the input pattern arrays.

12. The system of claim 1 wherein the same time point is represented in multiple estimate matrices.

13. The system of claim 1 wherein the prognostic module is configured to use the most recent estimate matrix to update the estimate values for use to determine the condition of the object.

14. The system of claim 1 wherein the prognostic module is configured to provide values for a single estimate vector to represent a single time point across multiple estimate matrices.

15. The system of claim 14 wherein the single estimate vector is an average, a weighted average, or a weighted norm of all of the estimate vectors at the single time point.

16. The system of claim 1 wherein the prognostic module is configured to provide values for a single estimate vector to represent each estimate matrix.

17. The system of claim 16 wherein the single estimate vector is an average, weighted average, or weighted norm of the estimate vectors within the estimate matrix.

18. The system of claim 1 wherein the prognostic module is configured to form a trend line for at least one parameter represented by the inferred estimate values to indicate the expected behavior of the object.

19. The system of claim 18 wherein the prognostic module is configured to form a new trend line with each new estimate matrix.

20. The system of claim 18 wherein the prognostic module is configured to form boundary trend lines to define a range of expected behavior of the object.

21. The system of claim 19 wherein the prognostic module is configured to form an upper boundary trend line with maximum estimate values from the time points, and a lower boundary trend line with minimum estimate values from the time points.