Method for analysis of the operation of a gas turbine

• US 20100262401A1
• Filed: 09/19/2008
• Published: 10/14/2010
• Est. Priority Date: 10/26/2007
• Status: Active Grant

• Current Assignee: Siemens Energy Global GmbH & Co. KG (Siemens AG)
• Original Assignee: Siemens AG
• Inventors: Pfeifer, Uwe, Sterzing, Volkmar
• Granted Patent: US 8,396,689 B2
• Time in Patent Office: Days
• US Class Current: 702/182
• CPC Class Codes: F01D 17/08 responsive to condition of working-fluid, e.g. pressure
• 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: F04D 27/001 Testing thereof; Determination or simulation of flow characteristics; Stall or surge detection, e.g. condition monitoring
• CPC Class Codes: F05D 2260/80 Diagnostics
• CPC Class Codes: F05D 2270/707 fuzzy logic
• CPC Class Codes: F05D 2270/709 with neural networks
• 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 method for analyzing the operation of a gas turbine is provided. A neural network based upon a normal operation of the gas turbine is learned. A dynamic pressure signal is read by a pressure sensor in or on the compressor of the turbine, and an operating parameter is read by a further sensor. The dynamic pressure signal is subjected to a frequency analysis, a parameter of a frequency spectrum of the pressure signal being determined. Based upon the measured operating parameter and the parameter of the frequency spectrum of the pressure signal, the neural network is learned. The measured operating parameter and the parameter of the frequency spectrum are input parameters, and a diagnostic characteristic value representing a probability of a presence of operation of the gas turbine as a function of the input parameters is output.

Claims

1-14. -14. (canceled)

15. A method of analyzing operation of a multistage compressor of a gas turbine including a plurality of compressor stages, wherein a neural network is trained based upon normal operation of the gas turbine, comprising:

measuring a dynamic pressure signal by a pressure sensor in or on the multistage compressor;

measuring an operating parameter of the gas turbine by a further sensor during normal operation of the gas turbine;

performing a frequency analysis of the dynamic pressure signal, wherein a parameter of a frequency spectrum of the pressure signal is determined;

training a neural network based upon the measured operating parameter and the parameter of the frequency spectrum of the pressure signal,

wherein the measured operating parameter and the parameter of the frequency spectrum of the pressure signal are input variables, and

wherein a diagnosis characteristic value representing a probability measure of a presence of normal operation of the gas turbine as a function of the input variables is an output variable;

determining a characteristic frequency band based upon a rotation speed of the gas turbine and a number of guide vanes and rotor blades in the relevant compressor stage as a parameter of the frequency spectrum for each compressor stage; and

calculating for each characteristic frequency band an energy component of the pressure signal contained therein, the energy component being used as input variable for the neural network.

16. The method as claimed in claim 15, further comprising:

providing a dynamic pressure signal and an operating parameter of the gas turbine which were measured during normal operation of the gas turbine.

17. The method as claimed in claim 15, wherein the energy component of the pressure signal is selected from the group consisting of a root mean square value,

an amplitude maximum,

a plurality of adjacent amplitude maxima of frequency lines within the characteristic frequency band, and

a combination thereof.

18. The method as claimed in claim 15, wherein the neural network is a radial basis function network.

19. The method as claimed in claim 15, wherein the diagnosis characteristic value is a confidence value normalized in a value range between 0 and 1 and representing the probability that a respective parameter combination of input variables is a parameter combination known during normal operation of the gas turbine.

20. The method as claimed in claim 15, wherein, as a further parameter of the frequency spectrum, a ratio of the energy component of a characteristic frequency band to the energy components of harmonics of the characteristic frequency band is taken into account for use as input variables for the neural network.

21. The method as claimed in claim 15,

wherein for each compressor stage a neural network is trained, each neural network including as input variables parameters of the frequency spectrum which are associated with a characteristic frequency band, and including as an output variable a diagnosis characteristic value, and

wherein the diagnosis characteristic value represents a probability measure for the presence of normal operation of the respective compressor stage as a function of the input variables.

22. The method as claimed in claim 21, further comprising:

determining an overall diagnosis characteristic value of the diagnosis characteristic values of the respective compressor stages based upon predetermined rules.

23. The method as claimed in claim 22, wherein the predetermined rules are fuzzy rules.

24. The method as claimed in claim 15, wherein the frequency analysis which is performed for the dynamic pressure signal comprises a fast Fourier transformation.

25. The method as claimed in claim 15, wherein the operating parameter to be measured is selected from the group consisting of

rotation speed of the gas turbine,

load on the gas turbine,

ambient pressure,

ambient temperature,

air humidity,

position of the guide vanes on the compressor in the gas turbine, and

a combination thereof.

26. The method as claimed in claim 16, wherein the operating parameter to be provided is selected from the group consisting of

rotation speed of the gas turbine,

load on the gas turbine,

ambient pressure,

ambient temperature,

air humidity,

position of the guide vanes on the compressor in the gas turbine, and

a combination thereof.

27. The method as claimed in claim 25, wherein the measured operating parameter and the dynamic pressure signal originate from normal operation of the gas turbine at a fixed rotation speed for different loads.

28. The method as claimed in claim 25, wherein the measured operating parameter and the dynamic pressure signal originate from normal operation of the gas turbine at a fixed rotation speed for different positions of the guide vanes.

29. The method as claimed in claim 26, wherein the measured operating parameter and the dynamic pressure signal originate from normal operation of the gas turbine at a fixed rotation speed for different loads.

30. The method as claimed in claim 26, wherein the measured operating parameter and the dynamic pressure signal originate from normal operation of the gas turbine at a fixed rotation speed for different positions of the guide vanes.

31. A method for monitoring a gas turbine comprising a neural network, the method comprising:

measuring a dynamic pressure signal by a pressure sensor in or on a compressor of the gas turbine;

measuring an operating parameter of the gas turbine by a further sensor during normal operation of the gas turbine;

performing a frequency analysis of the dynamic pressure signal, wherein a parameter of a frequency spectrum of the pressure signal is determined;

supplying the measured operating parameter and the parameter of the frequency spectrum of the pressure signal as input variables to the neural network; and

outputting a diagnosis characteristic value as output variable from the neural network.

32. The method as claimed in claim 31, further comprising:

outputting a warning signal when the diagnosis characteristic value is outside a predetermined value range.

33. An apparatus for monitoring a gas turbine, comprising:

a pressure sensor for measuring a dynamic pressure signal in or on a compressor of the gas turbine;

a further sensor for measuring an operating parameter of the gas turbine during operation of the gas turbine;

a frequency analysis device configured to perform a frequency analysis of the dynamic pressure signal, wherein, as a result, a parameter of a frequency spectrum of the pressure signal is determined; and

a neural network, the measured operating parameter and the parameter of the frequency spectrum of the pressure signal being supplied to the neural network as input variables, wherein the neural network outputs a diagnosis characteristic value as output variable.

34. The apparatus as claimed in claim 33, wherein the neural network is a radial basis function network.