Systems and methods related to detecting blowout occurrences in gas turbines

Patent metadata

Field	Value
Publication number	US-10352825-B2
Application number	US-201715592553-A
Country	US
Kind code	B2
Filing date	May 11, 2017
Priority date	May 11, 2017
Publication date	Jul 16, 2019
Grant date	Jul 16, 2019

Abstract

A system for controlling an operation of a combustor in a gas turbine that includes: an acoustic sensor configured to periodically measure a pressure of the combustor and generate a raw data stream having the pressure data points resulting from the periodic measurements; and a blowout detection unit configured to receive the raw data stream from the acoustic sensor. The blowout detection unit may include a processor and a machine-readable storage medium on which is stored instructions that cause the processor to execute a procedure related to a detection of a blowout precursor. The procedure may include an ensemble approach in which the detection of the blowout precursor depends upon a outcomes generated respectively by separate detection analytics.

First claim

The invention claimed is:

1. A system for controlling an operation of a combustor in a gas turbine, the system comprising:
the combustor;
an acoustic sensor configured to periodically measure a pressure of the combustor and generate a a raw data stream that comprises pressure data points resulting from the periodic measurement of the pressure; and
a blowout detection unit configured to receive the raw data stream from the acoustic sensor;
wherein the blowout detection unit comprises a processor and a machine-readable storage medium on which is stored instructions that cause the processor to execute a procedure related to a detection of a blowout precursor; and
wherein the procedure comprises an ensemble approach in which the detection of the blowout precursor depends upon a plurality of outcomes generated respectively from a plurality of separate detection analytics, the plurality of separate detection analytics of the ensemble approach comprising at least two selected from the following detection analytics:
a first detection analytic that comprises a calculation as to whether a selected pressure data point of the pressure data points within the raw data stream varies from a steady state condition, wherein the steady state condition is defined by the pressure data points of the raw data stream that fall within a look-back period that precedes the selected pressure data point;
a second detection analytic that comprises a calculation of a standard deviation for the pressure data points of the raw data stream within the look-back period;
a third detection analytic that comprises calculating an entropy value related to an entropy feature of the pressure data points of the raw data stream within the look-back period; and
a fourth detection analytic that comprises a manifold anomaly detection approach that includes calculating a value of an anomaly indicator that is based on a pattern formed when the pressure data points of the raw data stream within the look-back period are projected onto a manifold shape.

2. The system according to claim 1 , wherein the blowout precursor comprises an indication of an increased risk of a flame blowing out in the combustor;
wherein the look-back period comprises a sliding temporal window that is defined relative to the selected pressure data point;
wherein the look-back period comprises a period of between 1 and 20 minutes; and
wherein the selected pressure data point comprises a most current one of the pressure data points received by the blowout detection unit from the acoustic sensor.

3. The system according to claim 1 , wherein the look-back period comprises a period of between 8 and 12 minutes; and
wherein:
the first detection analytic comprises calculating whether the selected pressure data point is outside of three standard deviations of a mean of the pressure data points that fall within the look-back period;
the second detection analytic comprises calculating whether a value for a standard deviation for the pressure data points that fall within the look-back period is equal to about zero;
the entropy feature of the third detection analytic comprises a permutation entropy, wherein the calculation of the entropy value of the permutation entropy is based on permutation patterns given a temporal ordering of values of the pressure data points that fall within the look-back period; and
the projection of the manifold anomaly detection approach o the fourth detection analytic comprises a kernel-based Laplacian projection.

4. The system according to claim 3 , wherein the outcome of each of the first detection analytic, the second detection analytic, the third detection analytic, and the fourth detection analytic comprises one of: a positive result, which indicates an increased likelihood of the blowout precursor; and a negative result, which indicates a decreased likelihood of the blowout precursor;
wherein:
in regard to the first detection analytic, the positive result corresponds to the selected pressure data point being outside of the three standard deviations of the mean, and the negative result corresponds to the selected pressure data point being inside of the three standard deviations of the mean;
in regard to the second detection analytic, the positive result corresponds to the standard deviation not equaling zero, and the negative result corresponds to the standard deviation equaling zero;
in regard to the third detection analytic, the positive result corresponds to the entropy value exceeding a predetermined threshold for the entropy value, and negative result corresponds to the entropy value not exceeding the predetermined threshold for the entropy value; and
in regard to the fourth detection analytic, the positive result corresponds to the value of the anomaly indicator exceeding a predetermined threshold value of the anomaly indicator, and negative result corresponds to the value of the anomaly indicator not exceeding the predetermined threshold value for the anomaly indicator.

5. The system according to claim 4 , wherein the plurality of separate detection analytics of the ensemble approach comprises at least three selected from the first detection analytic; the second detection analytic; the third detection analytic; and the fourth detection analytic.

6. The system according to claim 4 , wherein the plurality of separate detection analytics of the ensemble approach comprises each of the first detection analytic; the second detection analytic; the third detection analytic; and the fourth detection analytic;
wherein the ensemble approach further comprises:
determining an overall ensemble result given the plurality of outcomes generated respectively from the plurality of separate detection analytics, the overall ensemble result being based on:
which of the first detection analytic, the second detection analytic, the third detection analytic, and the fourth detection analytic produces the positive result; and
which of the first detection analytic, the second detection analytic, the third detection analytic, and the fourth detection analytic produces the negative result;

basing the detection of the blowout precursor on the overall ensemble result.

7. The system according to claim 6 , wherein the step of determining the overall ensemble result comprises weighting the outcomes of the first detection analytic according to a relative predictive strength.

8. The system according to claim 7 , wherein a frequency of the periodic measurement of the pressure comprises at least 13,000 hertz and the raw data stream includes each of the pressure data points generated by the acoustic sensor;
wherein the procedure performed by the blowout detection unit further comprises:
sampling the raw data stream via a sampling process to generate a sampled raw data stream, wherein the blowout detection unit comprises both the raw data stream and the sampled raw data stream;
feeding the raw data stream to one or more of the first detection analytic, the second detection analytic, the third detection analytic, and the fourth detection analytic for use in determining the outcomes; and
feeding the sampled raw data stream to one or more of the first detection analytic, the second detection analytic, the third detection analytic, and the fourth detection analytic for use in determining the outcomes.

9. The system according to claim 8 , wherein the sampling process comprises selecting one of the pressure data points for every 100 to 200 of the pressure data points in the raw data stream;
wherein weight comprises the predictive strength of the outcomes of each of the third detection analytic and the fourth detection being greater than the predictive strength of the outcomes of each of the

Assignees

• Gen Electric

Inventors

• Andrews Peter Frank
• Caffrey Paul Oliver
• Reviere Robert Daniel

Classifications

• F23N2225/04

  pressure · CPC title

• F23N2241/20

  Gas turbines · CPC title

• G01L9/0008

  using vibrations · CPC title

• F23R2900/00013

  Reducing thermo-acoustic vibrations by active means · CPC title

• G01M15/14 Primary

  Testing gas-turbine engines or jet-propulsion engines · CPC title