Method for calculating confidence on prediction in fault diagnosis systems

• US 8,417,432 B2
• Filed: 04/30/2008
• Issued: 04/09/2013
• Est. Priority Date: 04/30/2008
• Status: Active Grant

• Current Assignee: Rtx Corporation
• Original Assignee: United Technologies Corporation (Raytheon Technologies Corporation d/b/a RTX)
• Inventors: Butler, Steven Wayne
• Primary Examiner(s): Tarcza, Thomas
• Assistant Examiner(s): PIPALA, EDWARD J
• Application Number:
  US12/112,021
• Publication Number:
  US 20090276136A1
• Time in Patent Office: 1,805 Days
• Field of Search: 701/1, 701/29, 701/30, 701/31, 701/34, 701/35, 701/29.1, 701/99, 701/100, 701/106, 701/30.2, 701/30.3, 701/30.8, 701/30.9, 701/31.1, 701/31.7, 701/31.8, 701/31.9, 701/32.1, 701/102, 701/103, 701/29.2, 702/179, 702181-185, 340/945, 340/963, 340/971, 244/53, 244/75.1, 244/76.R
• US Class Current: 701/100
• CPC Class Codes: F05B 2260/80 Diagnostics
• CPC Class Codes: G05B 23/0281 Quantitative, e.g. mathematical distance; Clustering; Neural networks; Statistical analysis

Abstract

A method and system is developed that provides a confidence measure of a prediction of a fault in a gas turbine engine. The confidence measure is developed based upon evaluating the results of a plurality of past predictions and comparing them to an actual fault.

Claims

1. A fault diagnosis system comprising:

a fault detection system for taking in a plurality of sensor readings from a gas turbine engine;

said plurality of sensors providing readings to a fault isolation system, said fault isolation system predicting a current fault based upon said readings, and to a network communicating with a database of prior predictions, and the database associating the prior predictions with a confidence level for the current prediction; and

said fault isolation system providing the predicted current fault to an output, and the confidence level in the current prediction also being provided to the output.

2. The system as set forth in claim 1, wherein the database is developed, at least in part, using a model-based fault isolation system.

3. The system as set forth in claim 2, wherein a regression network is trained based upon the model-based fault isolation system to provide the confidence level.

4. The system as set forth in claim 3, wherein said regression network is trained by calculating a linear distance between shift magnitudes received from a fault detection system and each of the data points received from the database, using the following equation:

5. The system as set forth in claim 2, wherein the database is also developed based upon actual examples of correct and incorrect predictions.

6. The system as set forth in claim 1, wherein the database is developed based upon actual examples of correct and incorrect predictions.

7. The system as set forth in claim 1, wherein the provided confidence level is in the form of a percentage chance that the predicted fault is an accurate fault.

8. The system as set forth in claim 7, wherein a bias factor is utilized to reduce the confidence level percentage if there are few or no previously encountered data points which closely resemble the current data point.

9. The system as set forth in claim 1, wherein the database associates the prior predictions with said confidence level at least in part utilizing the accuracy of said prior predictions.

10. The system as set forth in claim 1, wherein said output is a computer.

11. A method of operating a fault prediction system including the steps of:

(a) taking in a plurality of sensor readings from a gas turbine engine;

(b) providing readings to a fault isolation system, and predicting a current fault based upon said readings to a network communicating with a database of prior predictions, the database associating the prior predictions with a confidence level in the current prediction; and

(c) said fault isolation system providing the predicted current fault to an output, and the confidence level in the current prediction also being provided to the output.

12. The method as set forth in claim 11, wherein the database is developed, at least in part, using a model-based fault isolation system.

13. The method as set forth in claim 12, wherein a regression network is trained based upon the model-based fault isolation system to provide the confidence level.

14. The method as set forth in claim 13, wherein said regression network is trained by calculating a linear distance between shift magnitudes received from a fault detection system and each of the data points received from the database, using the following equation:

15. The method as set forth in claim 12, wherein the database also is developed utilizing actual examples of correct and incorrect predictions.

16. The method as set forth in claim 11, wherein the database is developed utilizing actual examples of correct and incorrect predictions.

17. The method as set forth in claim 11, wherein the provided confidence level is in the form of a percentage chance that the predicted fault is an accurate fault.

18. The method as set forth in claim 17, wherein a bias factor is utilized to reduce the confidence level percentage if there are few or no previously encountered data points which closely resemble the current data point.

19. The method as set forth in claim 11, wherein the database associates the prior predictions with said confidence level at least in part utilizing the accuracy of said prior predictions.

20. The method as set forth in claim 11, wherein the output is a computer.