System and method for anomaly detection

• US 9,752,960 B2
• Filed: 11/22/2011
• Issued: 09/05/2017
• Est. Priority Date: 11/22/2011
• Status: Active Grant

• Current Assignee: Electric Power Research Institute Incorporated
• Original Assignee: Electric Power Research Institute Incorporated
• Inventors: Angello, Leonard, Lieuwen, Tim, Noble, David Robert, Poole, Brian
• Primary Examiner(s): Do, An
• Application Number:
  US14/359,539
• Publication Number:
  US 20150051847A1
• Time in Patent Office: 2,114 Days
• Field of Search: 702 33, 702 35, 60772, 60773
• CPC Class Codes: F01D 21/003 Arrangements for testing or measuring for measuring vibrations G01H
• CPC Class Codes: F05D 2260/80 Diagnostics
• CPC Class Codes: F05D 2270/304 Spool rotational speed
• CPC Class Codes: F05D 2270/334 Vibration measurements
• CPC Class Codes: F05D 2270/708 with comparison tables
• CPC Class Codes: F23N 2241/20 Gas turbines
• CPC Class Codes: F23N 5/242 using electronic means
• CPC Class Codes: F23R 2900/00013 Reducing thermo-acoustic vibrations by active means
• CPC Class Codes: G01M 15/14 Testing gas-turbine engines or jet-propulsion engines

Abstract

A system and method for anomaly detection is provided. The system (100) and method (200) include utilizing one or more processors, such as in a server, for receiving (108) operational and dynamics data (104) from sensors associated with devices (102), filtering the data, establishing a set of baseline dynamics data and eliminating data dependencies (110). The system and method further include generating an expected level of data variation (112), identifying an anomaly based on a deviation of the device data from the baseline data normalized by the expected level of data variation (114), optionally correlating an anomaly with potential causes, and providing an output indicating an anomaly (116).

Claims

1. A computer implemented method for anomaly detection, the method comprising:

utilizing one or more processors and associated memory storing one or more programs for execution by the one or more processors, the one or more programs including instructions for;

receiving operational and dynamics data from a plurality of sensors associated with a plurality of devices;

filtering the data, wherein said filtering the data comprises data smoothing and eliminating outlying data, the data smoothing comprising;

Fourier transforming the dynamics data;

extracting an amplitude and frequency for a plurality of pressure oscillation spectral bins over a time period; and

calculating a backward running average;

establishing a set of baseline dynamics data including calculating a reference mean for each dynamic based on identifying historical data values

a. relating to a sliding time window, or

b. corresponding to a database query establishing reference data requirements and tolerances;

identifying previous points in time that satisfy said data requirements and tolerances; and

averaging the value of dynamics data related to said identified points in time;

eliminating data dependencies;

generating an expected level of data variation;

identifying an anomaly based on a deviation of the device data from the baseline data normalized by the expected level of data variation;

optionally correlating the anomaly with potential causes; and

providing an output indicating the anomaly.

2. The method of claim 1, wherein the devices comprise gas turbine combustor cans.

3. The method of claim 2, wherein the dynamics data comprise pressure oscillations.

4. The method of claim 2, wherein the dynamics data further comprise at least one of electromagnetic radiation, chemiluminescence, velocity oscillations or another observable related to combustion dynamic data.

5. The method of claim 1, wherein the time period comprises ten seconds to two hours.

6. The method of claim 1, wherein the spectral bins comprise a plurality of dynamics, optionally at least one of low frequency dynamics, first intermediate dynamics, second intermediate dynamics or high frequency dynamics.

7. The method of claim 1, wherein said eliminating outlying data comprises:

eliminating bad sensor data; and

/or

eliminating data excursions present in data related to a gas turbine combustor can but not present in data relating to adjacent gas turbine combustor cans.

8. The method of claim 1, wherein the sliding time window looks back at predefined time periods and the baseline data is updated at each pre-defined time period.

9. The method of claim 1, wherein said data requirements comprise at least one of power, machine-on time, fuel composition, transient operation, ambient temperature, inlet guide vane angle, exit temperature, fuel splits or flashback temperature.

10. The method of claim 1, wherein said eliminating data dependencies comprises:

subtracting the reference mean from the dynamics data for each device;

calculating the non-uniformity of each device relative to other devices; and

analyzing changes in one device relative to changes in the other devices.

11. The method of claim 1, wherein said expected level of data variation comprises:

quantifying a level of data variation across devices.

12. An anomaly detection system for detecting anomalies in a turbine engine, the anomaly detection system comprising:

an input data module configured to receive sensor data from the turbine engine;

a processing module adapted to

i) filter the data, wherein said filtering the data comprises data smoothing and eliminating outlying data, the data smoothing comprising;

Fourier transforming the dynamics data;

extracting an amplitude and frequency for a plurality of pressure oscillation spectral bins over a time period; and

calculating a backward running average;

ii) establish a set of baseline dynamics data including calculating a reference mean for each dynamic based on identifying historical data values a) relating to a sliding time window, or b) corresponding to a database query capable of establishing reference data requirements and tolerances, identifying previous points in time that satisfy said data requirements and tolerances and averaging the value of dynamics data related to said identified points in time,

iii) eliminate data dependencies,

iv) generate an expected level of data variation; and

v) identify an anomaly based on a deviation of the sensor data from the baseline data normalized by the expected level of data variation;

a database capable of storing sensor data and communicating with the processing module;

an output data module capable of reporting results identified by the processing module;

an interface module capable of communicating results reported by the output data module;

a processor capable of managing operation of the data input module, processing module, database, output data module and/or interface module; and

memory capable of storing instructions and data for execution by the system.

13. The system of claim 12, wherein said sliding time window looks back at predefined time periods.

14. The system of claim 13, wherein the processing module is further adapted to:

subtract the reference mean from the dynamics data for each device;

calculate the non-uniformity of each device relative to other devices; and

analyze changes in one device relative to changes in the other devices.

15. A non-transitory computer-readable storage medium on which is encoded executable program code for performing a method comprising:

receiving operational and dynamics data from a plurality of sensors associated with a plurality of devices;

filtering the data, wherein said filtering the data comprises data smoothing and eliminating outlying data, the data smoothing comprising;

Fourier transforming the dynamics data;

extracting an amplitude and frequency for a plurality of pressure oscillation spectral bins over a time period; and

calculating a backward running average;

establishing a set of baseline dynamics data including calculating a reference mean for each dynamic based on identifying historical data values

a. relating to a sliding time window, or

b. corresponding to a database query establishing reference data requirements and tolerances;

identifying previous points in time that satisfy said data requirements and tolerances; and

averaging the value of dynamics data related to said identified points in time;

eliminating data dependencies;

generating an expected level of data variation;

identifying an anomaly based on a deviation of the device data from the baseline data normalized by the expected level of data variation;

optionally correlating said anomaly with potential causes; and

providing an output indicating the anomaly.

16. The non-transitory computer-readable storage medium of claim 15, wherein said devices comprise gas turbine combustor cans and said dynamics data comprise pressure oscillations.

17. The non-transitory computer-readable storage medium of claim 16, wherein said eliminating data dependencies comprises:

subtracting the reference mean from the dynamics data for each device;

calculating the non-uniformity of each device relative to other devices; and

analyzing changes in one device relative to changes in the other devices.