Transient fault detection system and method using Hidden Markov Models

• US 20040176879A1
• Filed: 03/07/2003
• Published: 09/09/2004
• Est. Priority Date: 03/07/2003
• Status: Active Grant

• Current Assignee: Honeywell International Inc.
• Original Assignee: Honeywell International Inc.
• Inventors: Nwadiogbu, Emmanuel O., Menon, Sunil K.
• Granted Patent: US 6,868,325 B2
• Time in Patent Office: Days
• US Class Current: 701/1
• CPC Class Codes: F02C 9/00 Controlling gas-turbine plants; Controlling fuel supply in air- breathing jet-propulsion plants controlling air intakes F02C7/057; controlling turbines F01D; controlling compressors F04D27/00; controlling in general G05
• CPC Class Codes: G05B 23/021 adopting a different treatment of each operating region or a different mode of the monitored system, e.g. transient modes; different operating configurations of monitored system
• 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 11/2257 using expert systems

Abstract

A transient fault detection system and method is provided that facilitates improved fault detection performance in transient conditions. The transient fault detection system provides the ability to detect symptoms of engine faults that occur in transient conditions. The transient fault detection system includes a Hidden Markov Model detector that receives sensor data during transient conditions and determines if a fault has occurred during the transient conditions. Detected faults can then be passed to a diagnostic, system where they can be passed as appropriate to maintenance personnel.

Claims

1. A transient fault detection system for detecting transient faults in a turbine engine, the transient fault detection system comprising:

a Hidden Markov Model detector, the Hidden Markov Model receiving turbine sensor data from the turbine engine during a transient condition and analyzing the sensor data to determine a likelihood that a fault occurred during the transient condition.

2. The system of claim 1 wherein the Hidden Markov Model detector comprises a left-right Hidden Markov Model.

3. The system of claim 1 wherein the Hidden Markov Model detector includes a plurality of states, and wherein the plurality of states is equal in number to a number of samples in the turbine sensor data during the transient condition.

4. The system of claim 1 wherein the Hidden Markov Model is trained using a test data set, and wherein the test data set includes data from a normal transient condition, such that the Hidden Markov Model detector compares the turbine sensor data to the normal transient condition represented by the test data set.

5. The system of claim 1 wherein the Hidden Markov Model is trained using a test data set, and wherein the test data set includes data from an abnormal transient condition, such that the Hidden Markov Model detector compares the turbine sensor data to the abnormal transient condition represented by the test data set.

6. The system of claim 1 wherein the sensor data comprises engine speed.

7. The system of claim 1 wherein the sensor data comprises exhaust gas temperature.

8. The system of claim 1 wherein the transient condition comprises turbine engine startup.

9. The system of claim 1 wherein the transient condition comprises a change in engine power setting or engine operating mode.

10. The system of claim 1 wherein the Hidden Markov Model detector includes a plurality of states, with each state including a model output, and wherein the number of states and the number of model outputs is equal in number to a number of samples in the turbine sensor data during the transient condition, and wherein the turbine sensor data includes both engine speed data and exhaust gas temperature data, and wherein the Hidden Markov Model detector compares the turbine sensor data from the transient condition to a sample) transient condition represented by a test data set used to train the Hidden Markov Model detector.

11. An apparatus comprising:

a) a processor;

b) a memory coupled to the processor;

c) a transient fault detection program residing in the memory and being executed by the processor, the transient fault detection program including;

a Hidden Markov Model detector, the Hidden Markov Model receiving turbine sensor data from the turbine engine during a transient condition and analyzing the sensor data to determine a likelihood that a fault occurred during the transient condition.

12. The apparatus of claim 11 wherein the Hidden Markov Model detector comprises a left-right Hidden Markov Model.

13. The apparatus of claim 11 wherein the Hidden Markov Model detector includes a plurality of states, and wherein the plurality of states is equal in number to a number of samples in the turbine sensor data during the transient condition.

14. The apparatus of claim 11 wherein the Hidden Markov Model is trained using a test data set, and wherein the test data set includes data from a normal transient condition, such that the Hidden Markov Model detector compares the turbine sensor data to the normal transient condition represented by the test data set.

15. The apparatus of claim 11 wherein the Hidden Markov Model is trained using a test data set, and wherein the test data set includes data from an abnormal transient condition, such that the Hidden Markov Model detector compares the turbine sensor data to the abnormal transient condition represented by the test data set.

16. The apparatus of claim 11 wherein the sensor data comprises engine speed.

17. The apparatus of claim 11 wherein the sensor data comprises exhaust gas temperature.

18. The apparatus of claim 11 wherein the transient condition comprises turbine engine startup.

19. The apparatus of claim 11 wherein the transient condition comprises a change in engine power setting or engine operating mode.

20. The apparatus of claim 11 wherein the Hidden Markov Model detector includes a plurality of states, with each state including a model output, and wherein the number of states and the number of model outputs is equal in number to a number of samples in the turbine sensor data during the transient condition, and wherein the turbine sensor data includes both engine speed data and exhaust gas temperature data, and wherein the Hidden Markov Model detector compares the turbine sensor data from the transient condition to a sample transient condition represented by a test data set used to train the Hidden Markov Model detector.

21. A program product comprising:

a) a transient fault detection program, the transient fault detection program including;

a Hidden Markov Model detector, the Hidden Markov Model receiving turbine sensor data from the turbine engine during a transient condition and analyzing the sensor data to determine a likelihood that a fault occurred during the transient condition; and

b) signal bearing media bearing said program.

22. The program product of claim 21 wherein the signal bearing media comprises recordable media.

23. The program product of claim 21 wherein the signal bearing media comprises transmission media.

24. The program product of claim 21 wherein the Hidden Markov Model detector comprises a left-right Hidden Markov Model.

25. The program product of claim 21 wherein the Hidden Markov Model detector includes a plurality of states, and wherein the plurality of states is equal in number to a number of samples in the turbine sensor data during the transient condition.

26. The program product of claim 21 wherein the Hidden Markov Model is trained using a test data set, and wherein the test data set includes data from a normal transient condition, such that the Hidden Markov Model detector compares the turbine sensor data to the normal transient condition represented by the test data set.

27. The program product of claim 21 wherein the Hidden Markov Model is trained using a test data set, and wherein the test data set includes data from an abnormal transient condition, such that the Hidden Markov Model detector compares the turbine sensor data to the abnormal transient condition represented by the test data set.

28. The program product of claim 21 wherein the sensor data comprises engine speed.

29. The program product of claim 21 wherein the sensor data comprises exhaust gas temperature.

30. The program product of claim 21 wherein the transient condition comprises turbine engine startup.

31. The program product of claim 21 wherein the transient condition comprises a change in engine power setting or engine operating mode.

32. The program product of claim 21 wherein the Hidden Markov Model detector includes a plurality of states, with each state including a model output, and wherein the number of states and the number of model outputs is equal in number to a number of samples in the turbine sensor data during the transient condition, and wherein the turbine sensor data includes both engine speed data and exhaust gas temperature data, and wherein the Hidden Markov Model detector compares the turbine sensor data from the transient condition to a sample transient condition represented by a test data set used to train the Hidden Markov Model detector.

33. A method of detecting faults in transient conditions in a turbine engine, the method comprising the steps of:

a) receiving turbine sensor data from the turbine engine during a transient condition;

b) analyzing the sensor data with a Hidden Markov Model detector to determine a likelihood that a fault occurred during the transient condition.

34. The method of claim 33 wherein the Hidden Markov Model detector comprises a left-right Hidden Markov Model.

35. The method of claim 33 wherein the Hidden Markov Model detector includes a plurality of states, and wherein the plurality of states is equal in number to a number of samples in the turbine sensor data during the transient condition.

36. The method of claim 33 further comprising the step of training the Hidden Markov Model using a test data set, and wherein the test data, set includes data from a normal transient condition, such that the Hidden Markov Model detector compares the turbine sensor data to the normal transient condition represented by the test data set.

37. The method of claim 33 further comprising the step of training the Hidden Markov Model using a test data set, and wherein the test data set includes data from an abnormal transient condition, such that the Hidden Markov Model detector compares the turbine sensor data to the abnormal transient condition represented by the test data set.

38. The method of claim 33 wherein the sensor data comprises engine speed.

39. The method of claim 33 wherein the sensor data comprises exhaust gas temperature.

40. The method of claim 33 wherein the transient condition comprises turbine engine startup.

41. The method of claim 33 wherein the transient condition comprises a change in engine power setting or engine operating mode.

42. The method of claim 33 wherein the Hidden Markov Model detector includes a plurality of states, with each state including a model output, and wherein the number of states and the number of model outputs is equal in number to a number of samples in the turbine sensor data during the transient condition, and wherein the turbine sensor data includes both engine speed data and exhaust gas temperature data, and wherein the Hidden Markov Model detector compares the turbine sensor data from the transient condition to a sample transient condition represented by a test data set used to train the Hidden Markov Model detector.