Gas turbine engine anomaly detections and fault identifications

Patent metadata

Field	Value
Publication number	US-9818242-B2
Application number	US-201514934509-A
Country	US
Kind code	B2
Filing date	Nov 6, 2015
Priority date	Dec 16, 2014
Publication date	Nov 14, 2017
Grant date	Nov 14, 2017

Abstract

System and methods for detecting anomalies and identifying faults of a gas turbine engine may include a recorder in communication with a processor. The recorder may be configured to capture archival data of the gas turbine engine. A flight normalizer module may be configured to produce normalized results based on the archival data. A flight parameter features module may be configured to generate flight parameter features based on the normalized results. A data warehouse module may be configured to determine suspected fault classes by comparing the flight parameter features against training parameter features stored in the data warehouse module based on queries from the flight parameter features module. A majority vote module may be configured to determine a diagnosed fault class based on the suspected fault classes.

First claim

What is claimed is:

1. A system for detecting anomalies and identifying faults of a gas turbine engine, the system comprising:
a processor;
a recorder in communication with the processor, capturing from the gas turbine engine, by continuously sampling during operation of the gas turbine engine, data including one or more of internal temperature (T), pressure (P), and spool speeds;
a flight normalizer module executed by the processor, producing normalized results based on the data by calculating
θ= T inlet /T reference ,

δ= P inlet /P reference ,

P corrected =P /(θ a δ b )

P corrected Normalized =(100/( P corrected MAX −P corrected MIN ))*( P corrected −P corrected MIN ), and
wherein minimum and maximum values are expected minimum and maximum values for a nominal engine, calculated from engine simulations;

one of a weighted nearest neighbor (kNN) classification module executed by the processor or a flight parameter features module executed by the processor;
the weighted nearest neighbor (kNN) classification module executed by the processor performs a weighted nearest neighbor classification analysis, which finds mathematical relationships between n selected engine input parameters and m selected engine output parameters, by forming an n+m dimensional spatial index structure representing the input-output relationship at a sampled point, representing each time step in the stream of captured data, and evaluating resulting spatial relationships, and
wherein queries to the weighted nearest neighbor (kNN) classification module of the different points identifies whether the query points indicate normal or fault conditions, and, if a fault condition, an estimated time period for the fault;
the flight parameter features module executed by the processor generates flight parameter features based on the normalized results, by using a discrete cosine transform performed by a discrete cosine transform module or the weighted nearest neighbor (kNN) analysis performed by the weighted nearest neighbor classification module, to classify normalized data as nominal or type of fault;
a profile data module in communication with the processor, communicating nominal flight profile data sets, which include one or more of power level angle, altitude, mach, total air temperature used by an engine control, and engine condition data sets;
a test case generation module executed by the processor, generating a nominal condition engine class, representing a series of randomly distributed engine component performance levels that cover the expected level of engine-to-engine production or overhaul variation, by inputting combinations of the flight profile data sets, and the engine condition data sets into the engine control, and a corresponding plurality of fault classes by introducing a plurality of fault conditions into the flight profile data sets,
a training normalizer module executed by the processor, generating training normalized data in response to receiving and normalizing the nominal condition engine class, and the corresponding plurality of fault classes;
a training parameter features module executed by the processor, generating the training parameter features for distinguishing between normal conditions and fault classes in response to receiving the training normalized data, by using the discrete cosine transform performed by the discrete cosine transform module or the weighted nearest neighbor (kNN) analysis performed by the weighted nearest neighbor classification module;
a data warehouse module executed by the processor, determining suspected fault classes by comparing the flight parameter features against the training parameter features stored in the data warehouse module based on queries from the flight parameter features module, the queries performed using the weighted kNN nearest neighbor classification module; and
a majority vote module executed by the processor, determining a diagnosed fault class based on the suspected fault classes, by determining that the suspected fault class outputs correspond to a particular fault class a number of times above a particular threshold.

2. The system of claim 1 , wherein the training parameter features are generated by a training system that is in communication with the data warehouse module.

3. The system of claim 2 , wherein the training system includes an engine model driven by an engine control executed by the processor.

4. The system of claim 3 , further including a profile data module executed by the processor, the profile data module includes flight profile data sets and engine condition data sets.

5. The system of claim 4 , further including a test case generation module executed by the processor and configured to generate a nominal condition engine class by inputting combinations of the flight profile data sets and the engine condition data sets into the engine control.

6. The system of claim 5 , wherein the test case generation module is configured to generate a corresponding plurality of fault classes by introducing a plurality of fault conditions into the flight profile data sets.

7. The system of claim 6 , further including a training normalizer module executed by the processor and configured to generate training normalized data in response to receiving and normalizing the nominal condition engine class and the corresponding plurality of fault classes.

8. The system of claim 7 , further including a training parameter features module executed by the processor, the training parameter features module configured to generate the training parameter features in response to receiving the training normalized data.

9. The system of claim 1 , wherein the recorder is a continuous recorder.

10. The system of claim 1 , wherein the training parameter features and the flight parameter features are determined by using a discrete cosine transform module.

11. The system of claim 1 , wherein the training parameter features and the flight parameter features are utilized by a weighted kNN nearest neighbor classification module.

12. A method for detecting anomalies and identifying faults of a gas turbine engine using the system of claim 1 , the method comprising:
receiving data of the gas turbine engine from the recorder using a computer processor;
generating normalized data using the computer processor;
generating flight parameter features using the computer processor;
determining suspected fault classes by comparing the flight parameter features against training parameter features stored in a data warehouse module using the computer processor; and
determining a diagnosed fault class from the suspected fault classes using the computer processor.

13. The method of claim 11 , further including:
determining the flight parameter features and the training parameter features using a discrete cosine transform module.

14. The method of claim 12 , further including
using a weighted kNN nearest neighbor classification module to issue a query of the flight parameter features to the training parameter features.

15. The method of claim 12 , wherein
the training parameter features are generated from a training system in communication with the data warehouse module using the computer processor.

16. The method of claim 15 , wherein
the training system generates a nominal condition engine class and a plurality of fault classes using the computer processor.

17. The method of claim 15 , wherein
the training system normalizes the nominal condition engine class and the plurality of fault classes to generate tr

Assignees

• Univ Southern California
• Inha Univ

Inventors

• Volponi Allan J
• Geib Andrew F
• Gawecki Martin
• Kuo Chung Chieh
• Choi Wonik

Classifications

• G05B23/0281

  Quantitative, e.g. mathematical distance; Clustering; Neural networks; Statistical analysis · CPC title

• G07C5/0841

  Registering performance data (recording measured values G01D; information storage G11B) · CPC title

• G06N99/005

  Physics · mapped topic

• G05B23/0254

  based on a quantitative model, e.g. mathematical relationships between inputs and outputs; functions: observer, Kalman filter, residual calculation, Neural Networks · CPC title

• G06N5/047

  Pattern matching networks; Rete networks · CPC title