FAULT DETECTION SYSTEM AND METHOD USING MULTIWAY PRINCIPAL COMPONENT ANALYSIS

• US 20070124113A1
• Filed: 11/28/2005
• Published: 05/31/2007
• Est. Priority Date: 11/28/2005
• Status: Active Grant

• Current Assignee: Honeywell International Inc.
• Original Assignee: Honeywell International Inc.
• Inventors: Foslien, Wendy, Allumallu, Satya
• Granted Patent: US 7,243,048 B2
• Time in Patent Office: Days
• US Class Current: 702/185
• CPC Class Codes: G05B 23/024 Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks

Abstract

A fault detection system and method is provided that facilitates detection of faults that are manifest over a plurality of different operational phases. The fault detection system and method use multiway principal component analysis (MPCA) to detect fault from turbine engine sensor data. Specifically, the fault detection system uses a plurality of load vectors, each of the plurality of load vectors representing a principal component in the turbine engine sensor data from the multiple operational phases. The load vectors are preferably developed using sets of historical sensor data. When developed using historical data covering multiple operational phases, the load vectors can be used to detect likely faults in turbine engines. Specifically, new sensor data from the multiple operational phases is projected on to the load vectors, generating a plurality of statistical measures that can be classified to determine if a fault is manifest in the new sensor data.

Claims

1. A fault detection system for detecting faults in a plurality of turbine engines, the fault detection system comprising:

a multiway principal component analysis (MPCA) mechanism, the MPCA mechanism adapted to receive a batch of turbine sensor data from the plurality of turbine engines, the batch of turbine sensor data including sensor data from the plurality of turbine engines operating over a plurality of different operational phases, the MPCA mechanism including a plurality of load vectors, each of the plurality of load vectors representing a principal component in the turbine sensor data, the MPCA mechanism further adapted to project the batch of turbine sensor data onto the plurality of load vectors to generate a plurality of statistical measures; and

a classifier, the classifier adapted to receive the plurality of statistical measures and analyze the plurality of statistical measures to determine if a fault occurred in the plurality of turbine engines during the plurality of different operational phases, the classifier further generating an output indicating the fault responsive to the classifier determining that the fault occurred in the plurality of turbine engines during the plurality of different operational phases.

2. The system of claim 1 wherein the plurality of statistical measures comprise a plurality of scores.

3. The system of claim 1 wherein the plurality of statistical measures comprise a plurality of residual error estimates.

4. The system of claim 1 wherein the classifier is adapted to analyze the plurality of statistical measures by determining a distance between the plurality of statistical measures and statistical measures obtained from historical sensor data.

5. The system of claim 1 wherein the classifier is adapted to analyze the plurality of statistical measures by determining if the plurality of statistical measures lie within statistical confidence intervals obtained from historical sensor data.

6. The system of claim 1 wherein the classifier is adapted to analyze the plurality of statistical measures by determining clusters in which the plurality of statistical measures reside.

7. The system of claim 6 wherein the classifier is adapted to determine a temporal sequence of the clusters.

8. The system of claim 1 wherein the classifier is adapted to analyze the plurality of statistical measures by generating Q statistics from the plurality of statistical measures and determining if the Q statistics lie within statistical confidence intervals obtained from historical sensor data.

9. The system of claim 1 wherein the classifier is adapted to analyze the plurality of statistical measures by generating T² statistics from the plurality of statistical measures and determining if the T² statistics lie within statistical confidence intervals obtained from historical sensor data.

10. The system of claim 1 wherein the plurality of different operational phases includes ground acceleration, ascent to cruising altitude, and leveling off at cruising altitude.

11. The system of claim 1 wherein the batch of turbine sensor data is formatted into a three-dimensional matrix and wherein the MPCA mechanism is adapted to unfold the three-dimensional matrix into a two dimensional representation to facilitate projecting of the batch of turbine sensor data onto the plurality of load vectors.

12. A method of detecting faults in a turbine engine, the method comprising the steps of:

receiving a batch of turbine engine sensor data from a plurality of turbine engines, the batch of turbine engine sensor data including sensor data from the plurality of turbine engines operating over a plurality of different operational phases;

projecting the batch of turbine engine sensor data onto a plurality of multiway principal component analysis (MPCA) load vectors to generate a plurality of statistical measures;

classifying the plurality of statistical measures to determine if a fault occurred in the plurality of turbine engines during the plurality of different operational phases; and

generating an output indicating the fault responsive to determining the fault occurred in the plurality of turbine engines during the plurality of different operational phases.

13. The method of claim 12 wherein the plurality of statistical measures comprise a plurality of scores.

14. The method of claim 12 wherein the plurality of statistical measures comprise a plurality of residual error estimates.

15. The method of claim 12 wherein the step of classifying comprises determining a distance between the plurality of statistical measures and statistical measures obtained from historical sensor data.

16. The method of claim 12 wherein the step of classifying comprises determining if the plurality of statistical measures lie within statistical confidence intervals obtained from historical sensor data.

17. The method of claim 12 wherein the step of classifying comprises determining if the plurality of statistical measures create a cluster.

18. The method of claim 12 wherein the step of classifying comprises generating Q statistics from the plurality of statistical measures and determining if the Q statistics lie within statistical confidence intervals obtained from historical sensor data.

19. The method of claim 12 wherein the step of classifying comprises generating T² statistics from the plurality of statistical measures and determining if the T² statistics lie within statistical confidence intervals obtained from historical sensor data.

20. The method of claim 12 wherein the plurality of different operational phases includes ground acceleration, ascent to cruising altitude, and leveling off at cruising altitude.

21. The method of claim 12 further comprising the step of formatting the batch of turbine engine sensor data into a three-dimensional matrix and unfolding the three-dimensional matrix into a two dimensional representation prior to the step of projecting the batch of turbine engine sensor data onto a plurality of multiway principal component analysis (MPCA) load vectors to generate a plurality of statistical measures.

22. A program product comprising:

a) a fault detection program for detecting faults in a plurality of turbine engines, the fault detection program including;

a multiway principal component analysis (MPCA) mechanism, the MPCA mechanism adapted to receive a batch of turbine sensor data from the plurality of turbine engines, the batch of turbine sensor data including sensor data from the plurality of turbine engines operating over a plurality of different operational phases, the MPCA mechanism including a plurality of load vectors, each of the plurality of load vectors representing a principal component in the turbine sensor data, the MPCA mechanism adapted to project the batch of turbine sensor data onto the plurality of load vectors to generate a plurality of statistical measures; and

a classifier, the classifier adapted to receive the plurality of statistical measures and analyze the plurality of statistical measures to determine if a fault occurred in the plurality of turbine engines during the plurality of different operational phases, the classifier further generating an output indicating the fault responsive to the classifier determining that the fault occurred in the plurality of turbine engines during the plurality of different operational phases; and

b) computer-readable signal-bearing recordable media bearing said fault detection program.