Gas turbine operation analysis method

FIELD: machine building.

SUBSTANCE: at least one dynamic pressure signal is measured by means of at least one pressure sensor in or on the turbine compressor, as well as one or more operating parameters of the turbine are measured by means of one or more other sensors under normal operating conditions of the turbine, and/or dynamic pressure signal, as well as one or more other operating parameters, which have been measured under normal operating conditions of the turbine, are read out; at that, dynamic pressure signal is subject to frequency analysis, by means of which one or more parameters of frequency spectrum of pressure signal are determined. Based on one or more measured operating parameters and one or more parameters of frequency spectrum of pressure signal, one or more neutron networks are trained, which have one or more measured operating parameters and one or more parameters of frequency spectrum of pressure signal as input values, and have at least one diagnostics parameter as the output value, which represents the measure of probability for availability of normal operating conditions of the turbine depending on input values.

EFFECT: improving diagnostics accuracy of wear and damages to the turbine.

28 cl, 3 dwg

The invention relates to a method for analyzing the operation of the gas turbine, and method of controlling the operation of the gas turbine.

Modern gas turbines for power generation and industrial applications, as well as for aircraft engines include most often multistage axial compressors, which during operation is exposed to a variety of mechanisms of wear, contamination and other damage mechanisms that affect the operation of the compressor. Early recognition of such a state machine, deviating from the normal state, forms an essential prerequisite for the introduction of preventive maintenance activities to avoid critical operating States and unacceptable wear.

In the diagnosis and control of modern gas turbines are particularly important unambiguous classification and quantification of wear and damage. In particular, it is desirable that in gas turbines with multi-stage axial compressors can be accurately specify the compressor stage occurs sign of wear or damage and how strong it is or how far gone the damage in relation to the set maximum value. In addition, such methods of diagnosis and control galinafromodessa during normal operation and does not require stopping of the turbine.

In the prior art there are known various methods for diagnosing and monitoring for turbines. For example, in DE 4012278 A1 describes a diagnostic system for installation of a steam turbine with a neural network model. Using the model, the system can pre-study many data samples are dependent on the operational state of oscillation, so that when it occurs to generate an output signal indicating the operating status. For this purpose, and processed various forms of mechanical waves or acoustic waves, vibration or electromagnetic waves.

From US 2002/0013664 A1 it is known introduction classification for monitoring rotating components based on state machines. At the same time as possible input parameters may be involved pulsation pressure of the air compressor. Another method is known from US 7027953 B2.

These methods, as, for example, the method according to US 7027953 B2, in which for each of the observed speed compressor uses pressure sensors require a very large instrumental equipment in the form of a large number of sensors, and they can only detect severe damage, such as loss of the scapula. In addition, such methods in gas turbines with multi-stage compressors may not accurately correlate the damage with the compressor stage.

Ass is whose invention is a method of analysis and control operation of the gas turbine, with the help of a small number of sensors allows accurate diagnosis of the wear and damage of the turbine.

This problem is solved according to the independent claims. Further development of the invention is defined in dependent clauses.

According to the method corresponding to the invention, based on the normal operation of the gas turbine, study one or more neural networks. In this first measured dynamic pressure signal through at least one pressure sensor in or on, or with the compressor turbine, and the dynamic pressure signal means that is determined by the change in time of the pressure signal. The preferred frequency of sampling for the determination of the pressure signal are in the kHz range. Changes of pressure in the compressor occur during the passage of the blades past the vanes, which leads to the corresponding pressure fluctuations in the compressed air. Along with this, the dynamic pressure signal is also measured one or more operating parameters of the gas turbine by using other sensors. The method corresponding to the invention can thus be performed during operation of the turbine. If necessary, the dynamic pressure signal, as well as other operating parameters can be determined in advance and then to note the drop-in method, corresponding to the invention, are read, for example, from the data array.

The dynamic pressure signal according to the invention is subjected to frequency analysis, whereby you define one or more parameters of the frequency spectrum of the pressure signal. When it becomes known that for each stage of the compressor due to the interaction of the guide vanes and rotor blades in the compressor produces a periodic pressure fluctuations, which lead to a periodic signal that can be used to specify normal operation or deviating from it a working state.

Finally, based on the measured(s) work(s) parameter(s) and parameter(s) of the frequency spectrum of the pressure signal, attend one or more neural networks as input quantities measured include(s) work(s) parameter(s) and parameter(s) of the frequency spectrum of the pressure signal, and output values have at least one diagnostic indicator, which is a measure of the probability for the presence of normal operation of the gas turbine depending on the input values.

The method corresponding to the invention, characterized by the fact that, through the analysis of the dynamic pressure signal in combination with neural networks with a small number of dates is the ISR can be described by the normal operation of the compressor of the gas turbine. When this method is universally applicable to any gas turbines and only need first to be taught by measuring the operating parameters and the dynamic pressure signal of the considered gas turbine. In the subsequent control mode can then use neural networks a simple way to set the contrast of the studied normal mode or deviation from it so that when certain control operating parameters, including the dynamic pressure signal serves as the input variables in neural network.

The method corresponding to the invention is used for the multi-stage compressor of a gas turbine with a number of compressor stages, and in this case by means of frequency analysis as a parameter of the frequency spectrum for each compressor stage is determined by the characteristic frequency band, and for each characteristic frequency band is calculated, it contains a fraction of the energy of the signal pressure, in particular, as the root mean square (RMS) value. This energy is then used as an input value for one or more neural networks. The RMS value is well known in the art and is obtained by integrating the amplitudes associated with certain frequencies, the frequencies in the frequency band. The method corresponding to the invention, full-time is well-suited for multi-stage compressor, as the characteristics of the individual compressor stages are very well described by the corresponding frequency band, which is determined by a frequency analysis of the dynamic pressure signal. This method creates the opportunity to find work properly specifically for the individual compressor stages. Alternatively, or in addition to the RMS value, as input quantities can also be applied to the amplitude the maximum value of the frequency component or amplitude maximum value of the multiple adjacent frequency components characteristic of the frequency band.

The preferred way in method corresponding to the invention, applied the so-called network of radial basis functions (also called RBF-networks), which are known from the prior art. Can also be used for further development of such networks. These networks include many radial basis functions, such as Gaussian functions in the hidden layer, and the parameters of these Gaussian functions are taught. In the here described invention as a target parameter of the radial basis functions training is provided in relation to the likelihood that in the normal operation mode will have a combination of measured operating parameters and the dynamic signal DAVLENIYa document WO 99/48020 A2 describes the use of networks of radial basis functions in connection with the control of the rolling forces in the steel plant. The principles can likewise be transferred to appropriate the invention the analysis of the compressor of the gas turbine.

In another preferred form of execution of the method corresponding to the invention, at least one diagnostic indicator represents a confidence value, which is normalized in the range from 0 to 1 and represents the probability that the corresponding combination of parameters of the input values is known combination of parameters normal operation of the gas turbine. This method creates a simple representation of the metric diagnosis, and trust is close to 1, in particular, shows that there is a normal mode of operation of the gas turbine, and a confidence value less than 0.5 indicates that there was an unusual combination of parameters, which suggests that in the operation of the compressor has failed.

In another implementation of the method corresponding to the invention, as another parameter of the frequency spectrum for training the neural network can be considered the ratio of the share of energy characteristic of the frequency band to share the energy of higher harmonics of the characteristic frequency band.

When applying the method corresponding to the invention, in multi-stage compressors preference is sustained fashion for each compressor stage is trained neural network, moreover, the corresponding neural network as an input value is set in accordance with the parameters of the frequency spectrum related to the characteristic frequency band. This correspondence is obtained, thus, due to the characteristic frequency of the corresponding compressor stage, which is determined from the number of blades of the compressor stage and the actual number of revolutions of the gas turbine. Each neural network has a metric of diagnosis as the output value, and the diagnostic indicator is a measure of the probability of having a normal operation mode of the corresponding compressor stage depending on the input values. In this way, in the application of neural network for control of a gas turbine can be defined, in which the compressor stage is the failure of the operation. In another form of execution of the individual diagnostic indicators of compressor stages can be reduced to a common metric diagnosis, and this Association is based on certain rules, such as rule-based fuzzy logic or on the basis of discrete rules.

In a preferred form of execution of the method corresponding to the invention, for a frequency analysis of the dynamic pressure signal is applied the fast Fourier transform, trabulsi the small computation time, which converts the signal from the time domain into the frequency domain.

As the operating parameters, which, together with the dynamic pressure signal are determined in accordance with the invention, can be taken into consideration one or more of the following options:

- the number of revolutions of the gas turbine,

- load gas turbine

- ambient pressure,

- ambient temperature,

- the humidity of the air,

the position of the guide vanes on the compressor of the gas turbine.

Suitable normal mode of operation in which the relevant operating parameters and the dynamic pressure signal is preferably generated in such a way that during this mode of operation of the gas turbine is operated with a constant speed for various loads and/or provisions of the guide blades.

The above-described method, in which neural networks are trained on the basis of the normal operation of the gas turbine can be realized, in particular, as a computer program product. This computer program product includes program code stored on a machine-readable carrier for performing the method when the program is executed on a computer.

As stated above, trained according to the invention the neural network is then used to control the gas that is Bina to establish working conditions, deviating from the normal operation mode. Therefore, the invention also includes a method of controlling the gas turbine based on one or more neural networks trained according to the above method. This method of control is measured essentially the same magnitude as in the corresponding way of learning. Instead of training neural networks measured values are served now in the trained network as input quantities, and as a result receive the appropriate diagnostic indicator, which reproduces, what is the probability that a normal operation mode.

In particular, in the way of control during operation of the gas turbine are the following steps:

at least one dynamic pressure signal is measured, at least one pressure sensor in or on the compressor of the gas turbine, and, in addition, one or more operating parameters of the gas turbine is measured by other sensors;

the dynamic pressure signal is subjected to frequency analysis so determined one or more parameters of the frequency spectrum of the dynamic pressure signal;

- one or more measured operating parameters and one or more parameters of the frequency spectrum of the dynamic pressure signal serves as the input values in one or more of the scientists neural networks, and as the output values of one or more neural networks is given at least one diagnostic indicator.

Optionally, in the method of control may be issued a warning if one or more diagnostic indicators are outside a predefined range of values, that is, if by diagnostic indicator indicates that with high probability a condition deviating from normal operation of the gas turbine.

Along with the above-described control method, the invention also relates to a device for controlling a gas turbine, which is designed so that it can execute the above-described control method.

In particular, such a device includes the following components:

at least one pressure sensor for measuring at least one dynamic pressure signal in or on the compressor of the gas turbine, and one or more other sensors for measuring one or more operating parameters of the gas turbine during operation of the gas turbine;

device frequency analysis, whereby the dynamic pressure signal may be subjected to frequency analysis so determined one or more parameters of the frequency spectrum of the dynamic pressure signal;

one or more about the Chennai neural networks, which may be filed one or more measured operating parameters and one or more parameters of the frequency spectrum of the dynamic pressure signal as input quantities, and as output values can issue at least one diagnostic indicator.

The preferred way is the control device also includes means by which the above-described neural network learning can be done in normal mode.

The invention also relates to a gas turbine, which contains corresponding to the invention a device for controlling a gas turbine.

Examples of carrying out the invention hereinafter described based on the attached drawings showing the following:

Figure 1 - schematic representation of the method of controlling the operation of a gas turbine according to the form of execution of the invention;

2 is a diagram that represents the parameters of the frequency spectrum of the dynamic pressure signal is defined according to the form of execution of the invention; and

Figure 3 is a view that depicts the change over time in certain embodiment, the RMS values of the pressure signal of the gas turbine based on changing load and position of the guide blades.

Figure 1 shows a block diagram otobrajayuschii stages of the method forms the carrying out of the invention of the method of monitoring the operation of the gas turbine. In the form of execution according to Fig 1 is controlled by a gas turbine, the design of which is in itself known and is therefore described only briefly. The gas turbine includes a multistage axial compressor 2 working with multiple disks and rows of guide blades, and compressor made many compressor stages with guide vanes and rotor blades. By means of the guide vanes is set to the angle of air flow in the compressor for operating the blade, and the blade compresses and pumps the air further. For multi-stage axial compressor 2 in the turbine adjacent the combustion chamber, in which the appropriate fuel is burned using supplied through the air compressor, whereby the turbine is driven.

Shown in figure 1 turbine, a plurality of sensors that determine the appropriate operating parameters of the turbine. The sensor 4 is a sensor that measures ambient temperature and emits a corresponding measured signal V1. The sensor 5 is a pressure sensor that measures the air pressure in the environment and issues a corresponding measured signal V2. Reference position 6 marked the humidity sensor, which measures the humidity of the air and emits a corresponding measured signal V3. In addition, provided the n sensor 7, which compressor inlet measures the position of the adjustable guide vanes, and the vanes in a gas turbine can be changed by using the appropriate control device. The measured value of the position of the guide vanes in figure 1 is designated as V4.

Finally at the output of the compressor includes a pressure sensor 8, which dynamically measures the outlet pressure of the compressor in the form of a measured signal V5. When this "dynamically" means, as determined by the change in time of the sound pressure with an appropriate sampling rate, which is determined by time pressure. Measurement, in particular, is dynamic in that case, if the sampling frequency is in the kHz range and above. The measured pressure signal occurs due to the fact that in the individual compressor stages when working blade of the compressor passes by the guide vanes and thereby generate periodic pressure wave in the compressed air, and the period of the pressure wave depends on the number of guides and blades of the respective compressor stage. A certain dynamic pressure signal contains, therefore, in view of the many compressor stages, the set of periodic components.

Instead of applying only dates the ICA 8 pressure can also be used multiple pressure sensors, in particular, for measurements may involve existing pressure sensors, which in many turbines are used in the combustion zone to control the stability of combustion. Otherwise can be used to install a pressure sensor in the output cone or at the inlet of the compressor of the gas turbine. In accordance with the invention, the signal V5 first stage S1 is subjected to analog-to-digital (A/D) conversion and digital signal then at step S2 performs a fast Fourier transform (FFT) to determine the frequency spectrum of the signal. Performed on stage S2 FFT-transformation is so finely tuned to the frequency resulting from the number of revolutions of the gas turbine and the number of guides and blades that separate frequencies clearly separately can be correlated with the compressor stages. As the FFT-conversion gain characteristic of the frequency band with the corresponding amplitudes of the individual frequency.

Such a frequency spectrum is shown for example in the diagram in figure 2. This chart is referred to as the chart of Campbell (Campbell). The abscissa axis shows the frequency f in the pressure signal, and the y - axis the time t. The amplitude of certain frequencies in figure 2 has a color code, and because black and white represent this color coding is not visible. Usually red is the first color is used to indicate high amplitudes. For example, in figure 2 the observed range, which has a high amplitude. In the view of figure 2 at time t had to change certain operating parameters of the turbine. In particular, changes in load and position of the guide vanes of the compressor. From here turn out changing amplitude, and, however, the characteristic frequency band of the time remain the same. Especially clearly visible in figure 2 frequency polacy F1, F2, F3 and F4. Each of the frequency bands F1 to F4 is with this compressor stage axial compressor 2 of the gas turbine 1, that is, with each compressor stage when functioning with a certain number of revolutions corresponds to the frequency band with the characteristic frequency.

According to the step S3 of figure 1 then the evaluation is shown in figure 2 frequency bands, and, if necessary, corrects it using the appropriate model. At step S3, for each frequency band by integration over the frequency band is determined by the fraction of the energy of the pressure signal in the corresponding frequency band, with the proportion of the energy is given as the so-called root mean square (RMS) value. This RMS value is a parameter well known to the specialist. As a result of step S3 receive, thus, for each compressor stage features the terrorist RMS-value, moreover, in figure 1, for example, presents four RMS-values from R1 to R4 for the four compressor stages.

Figure 3 again shows the corresponding diagram is a temporary representation of the RMS-values in the turbine 20 of the characteristic frequency bands, which are correlated with the respective one compressor stage of the compressor of the gas turbine. Along the x-axis here is the put the time in seconds, and along the y - axis the corresponding RMS values of individual bands. The chart also has a color code, and each band is represented in a different color, which, however, due to the black-and-white representation of figure 3 is not visible. For individual frequency bands from band 1 to band 20 is shown in the diagram on the legend on the right. Along with the RMS-values chart in figure 3 contains also carried out during operation of the turbine load changes and the position of the guide vanes. This is shown by means of respective lines with the designation of "load" for load and "IGV" for vanes on the legend. For signature corresponding graph line temporary load change is indicated by L1, and the provisions of the guide vanes through L2. Individual values for load or position of the guide vanes when this is presented as a proportional value p is the tool of the ordinate on the right edge of the chart.

From figure 3 we can see that when you change the power, which is achieved by changing the mass flow of the compressor, get a very clear response to the RMS values, even with a slight change of mass flow. On the other hand, also shows that the complex responses of the system can be obtained depending on the operating condition. Using the RMS-values in combination with other operating parameters, of which figure 3 shows the load and the position of the guide vanes, it is possible in accordance with the invention to make a conclusion on deviating from the normal state of the gas turbine or compressor.

In order from the measured operating parameters, and RMS-values accordingly output a diagnostic indicator, in accordance with the invention are used in the neural network. In the described form of execution of the invention is applied neural model, preferably based on radial basis functions, which are also known as RBF network. The fundamental structure of such networks are well known in the prior art and therefore not explained here in detail. Such networks consist of a layer of input and output and study the parameters of the radial basis functions, such as Gaussian functions, based on the input values in the input layer, from here to approximate functional led the e and the distribution of input values. In the described form of execution for each compressor stage corresponding RBF network was trained on operating parameters, and the corresponding RMS values of the compressor stage, and the training was conducted on the basis of measurements during normal operation of the gas turbine.

A separate RBF network is produced as the output values are normalized between 0 and 1 trust value for a set of input values, i.e. existing in a particular time operating parameters and the corresponding RMS values, shows how high is the probability that in normal operation there will be such a combination of the RMS-values and operating parameters. The higher the confidence value, the more likely it is actually normal operation. In contrast, the small condence values mean that with high probability is working properly in the corresponding compressor of the gas turbine.

Accordingly trained neural network function as the approximative encapsulator data, and at step S4 of figure 1 receives as input values of the individual operating parameters according to the measured signals V1-V4 and RMS-values R1-R4. Step S4, for example, is divided into three podata S401, S402 and S403. At step S401 operating parameters and RMS-values of R1 and R2 is served in the corresponding neuron is diversified network of the respective compressor stage. At step S402 RMS-value R3 and the step S403 RMS-value of R4 is served in the neural network corresponding compressor stage. As a result of step S4 will receive for each neural network corresponding condence value in the range from 0 to 1. For trust values between 0.5 and 1 could be concluded that there is a normal mode, while for the trust value less than 0.5 is diagnosed, what is the mode of failures. These diagnostic values are given at step S5. When selecting operating parameters as input values for the neural networks is not required to separate parameters were unambiguous relationship. On the contrary, every distribution of combinations of parameters can be studied, if there is enough operational data for training the neural models. For high selectivity in the detection of unusual conditions is appropriate for all settings, it has a noticeable impact on the system, to use as input data for training the neural network.

Ultimately, the above described method for characteristic values of energy of frequency bands of each compressor stage one or more encapsulation data in the form of RBF networks are trained on a wide variety of different combinations of pairs is m, and trained encapsulator data are then used to control a gas turbine, in order to detect the operation mode failures. As parameters when training or control of a gas turbine is taken into account, in particular, the speed, the load, the position of the guide vanes, air pressure, ambient pressure, humidity, etc. These values along with the amplitudes of the energies of the characteristic frequencies are an important input variable of encapsulation data. Additionally it can be used the ratio of the RMS-values of the characteristic frequency to higher harmonics in the higher frequency bands. In addition, in one form of carrying out of the invention of the method of the condence values of the individual encapsulation data can be combined. Thus, for example, to determine the overall reliability availability normal mode. This can in particular be carried out on the basis of rules of the fuzzy logic or discrete rules that Express the known relations to behavior and interaction of individual compressor stages. In the result, you can use described here forms the carrying out of the invention of the method on the basis of measured values of a small number of pressure sensors to diagnose the quality and condition of the individual to the pressure-speed axial compressor of a gas turbine.

Using the appropriate the invention, a method provides several advantages. In particular, even when a slight amount of pressure sensors, for example, already with one pressure sensor for the entire compressor can be diagnosed condition of the compressor, which reduces the overall cost of the control operation of the gas turbine. In addition, corresponding to the invention the method can be simply aligned with the various gas turbines due to the fact that specifically for this gas turbine first mode of training the neural network are trained, and then on the basis of these trained networks is the monitoring of gas turbines. In addition, using the appropriate invention of the method allows rapid and high-frequency control of the compressor of the gas turbine during operation, and provides the possibility of obtaining long-term information for the service life of the gas turbine. In addition, can also be recognized by the hidden changes in relation to the normal operation of the gas turbine by determining a confidence value for a long time. Thereby, it is possible to reduce maintenance costs, since the invention through appropriate diagnostics damage detected in a timely manner, and thus may be relevant to discover the war damage repairs instead to carry out the repair with fixed intervals or purely preventive.

1. The method of analysis of the functioning of the multistage compressor (2) gas turbine (1) with a certain number of compressor stages,
when one or more neural networks are trained based on the normal operation of the gas turbine (1),
measured, at least one dynamic signal (V5) pressure by means of at least one sensor (8) pressure in or on the compressor (2) gas turbine (1), and measured one or more operating parameters (V1, V2, V3, V4) of the gas turbine (1) with one or more other sensors (4, 5, 6, 7) during normal operation of the gas turbine (1), and/or read dynamic signal (V5) pressure, as well as one or more operating parameters (V1, V2, V3, V4) of the gas turbine (1), which were measured during normal operation of the gas turbine (1);
dynamic signal (V5) pressure is subjected to frequency analysis, whereby you define one or more parameters of the frequency spectrum of the signal (V5) pressure;
based on one or more measured operating parameters (VI, V2, V3, V4) and one or more parameters of the frequency spectrum of the signal (V5) pressure, study one or more neural networks as input values have one or more measured Rabochaya (VI, V2, V3, V4) and one or more parameters of the frequency spectrum of the signal (V5) pressure, and as the output values have at least one diagnostic indicator, which is a measure of the probability for the presence of normal operation of the gas turbine (1) depending on the input values,
characterized in that the parameter of the frequency spectrum for each compressor stage is determined by the characteristic frequency band (F1, F2, F3, F4) on the basis of the number of revolutions of the gas turbine and the number of vanes and blades of the respective compressor stage, and for each characteristic frequency band is calculated, it contains a fraction of the energy of the signal (V5) pressure, in particular, RMS, and/or the amplitude maximum and/or multiple adjacent amplitude peaks of the frequency components within the characteristic frequency bands (F1, F2, F3, F4) for use as input values one or more neural networks.

2. The method according to claim 1, in which one or more neural networks are networks of radial basis functions.

3. The method according to claim 1 or 2, in which at least one diagnostic indicator represents a confidence value, which is normalized in the range from 0 to 1 and represents the probability that the corresponding whom inacia the parameters of the input values is known combination of parameters in the normal mode of operation of the gas turbine (1).

4. The method according to claim 1 or 2, in which as another parameter of the frequency spectrum takes into account the ratio of the share of energy characteristic of the frequency band (F1, F2, F3, F4) the shares of energy of higher harmonics of the characteristic frequency bands (F1, F2, F3, F4) for use as input values to one or more neural networks.

5. The method according to claim 3, in which as another parameter of the frequency spectrum takes into account the ratio of the share of energy characteristic of the frequency band (F1, F2, F3, F4) the shares of energy of higher harmonics of the characteristic frequency bands (F1, F2, F3, F4) for use as input values to one or more neural networks.

6. The method according to claim 1 or 2, in which for each compressor stage is trained neural network, and the corresponding neural network as input values has the parameters of the frequency spectrum related to the characteristic frequency band (F1, F2, F3, F4), and each neural network has a metric of diagnosis as the output value,
moreover, this diagnostic indicator is a measure of the probability of having a normal operation mode of the corresponding compressor stage depending on the input values.

7. The method according to claim 3, in which for each compressor stage is trained neural network, and the corresponding neural network as an input is elicin has parameters of the frequency spectrum, related to the characteristic frequency band (F1, F2, F3, F4), and each neural network has a metric of diagnosis as the output value,
moreover, this diagnostic indicator is a measure of the probability of having a normal operation mode of the corresponding compressor stage depending on the input values.

8. The method according to claim 5, in which for each compressor stage is trained neural network, and the corresponding neural network as input values has the parameters of the frequency spectrum related to the characteristic frequency band (F1, F2, F3, F4), and each neural network has a metric of diagnosis as the output value,
moreover, this diagnostic indicator is a measure of the probability of having a normal operation mode of the corresponding compressor stage depending on the input values.

9. The method according to claim 6, in which on the basis of certain rules, especially rule-based fuzzy logic is determined by the overall diagnostic value of the diagnostic indicators of the respective compressor stages.

10. The method according to claim 7 or 8, in which on the basis of certain rules, especially rule-based fuzzy logic is determined by the overall diagnostic value of the diagnostic indicators of the respective compressor the steps.

11. The method according to claim 1 or 2, in which a frequency analysis of the dynamic of the signal (V5) pressure is applied the fast Fourier transform.

12. The method according to claim 3, in which a frequency analysis of the dynamic of the signal (V5) pressure is applied the fast Fourier transform.

13. The method according to claim 4, in which a frequency analysis of the dynamic of the signal (V5) pressure is applied the fast Fourier transform.

14. The method according to any of pp.5, 7, 8 or 9, in which a frequency analysis of the dynamic of the signal (V5) pressure is applied the fast Fourier transform.

15. The method according to claim 10, in which a frequency analysis of the dynamic of the signal (V5) pressure is applied the fast Fourier transform.

16. The method according to claim 1 or 2, in which the operating parameters (VI, V2, V3, V4) are measured or read one or more of the following options:
the number of revolutions of the gas turbine (1),
the load of the gas turbine (1),
the ambient pressure,
ambient temperature,
humidity,
the position of the guide vanes on the compressor (2) gas turbine (1).

17. The method according to claim 3, in which the operating parameters (V1, V2, V3, V4) are measured or read one or more of the following options:
the number of revolutions of the gas turbine (1),
the load of the gas turbine (1),
the ambient pressure,
the temperature of the environment is s,
humidity,
the position of the guide vanes on the compressor (2) gas turbine (1).

18. The method according to any of pp.5, 7, 8, 9, 12, 13 or 15, in which the operating parameters (V1, V2, V3, V4) are measured or read one or more of the following options:
the number of revolutions of the gas turbine (1),
the load of the gas turbine (1),
the ambient pressure,
ambient temperature,
humidity,
the position of the guide vanes on the compressor (2) gas turbine (1).

19. The method according to claim 4, in which the operating parameters (V1, V2, V3, V4) are measured or read one or more of the following options:
the number of revolutions of the gas turbine (1),
the load of the gas turbine (1),
the ambient pressure,
ambient temperature,
humidity,
the position of the guide vanes on the compressor (2) gas turbine (1).

20. The method according to claim 6, in which the operating parameters (V1, V2, V3, V4) are measured or read one or more of the following options:
the number of revolutions of the gas turbine (1),
the load of the gas turbine (1),
the ambient pressure,
ambient temperature,
humidity,
the position of the guide vanes on the compressor (2) gas turbine (1).

21. The method according to clause 16, in which the measured operating parameters (V1, V2, V3, V4) and a dynamic signal (V5) pressure soo which correspond to the normal operation mode, wherein the gas turbine (1) is operated with a constant speed for various loads and/or provisions of the guide blades.

22. The method according to any of p, 19 or 20, in which the measured operating parameters (V1, V2, V3, V4) and a dynamic signal (V5) pressure correspond to the normal mode of operation in which the gas turbine (1) is operated with a constant speed for various loads and/or provisions of the guide blades.

23. The method according to p, in which the measured operating parameters (V1, V2, V3, V4) and a dynamic signal (V5) pressure correspond to the normal mode of operation in which the gas turbine (1) is operated with a constant speed for various loads and/or provisions of the guide blades.

24. The computer-readable medium containing stored thereon program code for performing the method according to any of the preceding paragraphs, when the program code is executed on the computer.

25. The control method of a gas turbine (1), based on one or more neural networks trained according to the method according to any one of claims 1 to 23, in which during operation of the gas turbine:
measured, at least one dynamic signal (V5) pressure, at least one sensor (8) pressure in or on the compressor (2) gas turbine (1), and measured one or more operating parameters (V1, V2, V3, V4) of the gas turbine (1) through other sensors (4, 5, 6, 7) during normal operation of the gas turbine (1);
dynamic signal (V5) pressure is subjected to frequency analysis so determined one or more parameters of the frequency spectrum of the signal (V5) pressure;
one or more measured operating parameters (V1, V2, V3, V4) and one or more parameters of the frequency spectrum of the signal (V5) pressure serves as input values in one or more trained neural networks, and as output the values of one or more neural networks is given at least one diagnostic indicator.

26. The method according A.25 that produces a warning signal if one or more diagnostic indicators are outside a predefined range of values.

27. Device for controlling a gas turbine (1)containing one or more trained neural networks implementing the method according to any one of claims 1 to 23, comprising:
at least one sensor (8) pressure for measuring at least one dynamic signal (V5) pressure in or on the compressor (2) gas turbine (1), and one or more other sensors for measuring one or more operating parameters (V1, V2, V3, V4) of the gas turbine (1) when the operation of the gas turbine (1);
the device frequency analysis, by which the dynamic signal (V5) pressure may be subjected to frequent Tomo analysis, thanks to determined one or more parameters of the frequency spectrum of the signal (V5) pressure;
one or more trained neural networks that were trained, and to which the input values may be filed one or more measured operating parameters (VI, V2, V3, V4) and one or more parameters of the frequency spectrum of the signal (V5) pressure, and which as output values can issue at least one diagnostic indicator.

28. Gas turbine containing the device according to item 27.