JP6927525B2 - Gas turbine abnormal condition detection device and abnormal condition detection method - Google Patents

Description

The present invention relates to an abnormal state detecting device for detecting the occurrence of an abnormal state of a gas turbine, and an abnormal state detecting method.

Conventionally, in gas turbines, dilute premixed combustion is known as a combustion method for reducing NOx generated in a combustor, but in the dilute premixed combustion, combustion vibration, blowout, etc. are performed. It is known that unstable physical phenomena occur. Furthermore, the flutter that the rotary blade vibrates when the flow velocity of the fresh air passing through the compressor exceeds a certain level, and the flow rate and pressure of the fresh air periodically increase as the flow rate of the fresh air passing through the compressor decreases. Unstable physical phenomena such as violently vibrating surges and turning stalls may occur. Here, fresh air includes the concept of both air alone and a mixture of air and fuel gas.
Since the unstable physical phenomena described above often occur suddenly, a technique capable of detecting the occurrence of these unstable physical phenomena in advance is desired.
For example, in the technique disclosed in Patent Document 1, time-series data of the pressure of a gas turbine combustor is used to detect and avoid a sign of sudden blowout in a gas turbine that performs dilute premixed combustion. Has been proposed as a method of performing a fast Fourier transform on the data and analyzing it in the frequency domain.

Japanese Unexamined Patent Publication No. 2006-70898

In the technique disclosed in Patent Document 1, in order to detect a sign of blowout, which is an unstable physical phenomenon, time-series data of the pressure of a gas turbine combustor is obtained by fast Fourier, which requires a huge amount of calculation. There is a problem that it is necessary to execute the calculation to be converted over time, a computer having a high calculation ability must be provided, and equipment cost is high. In addition, it is difficult to properly detect signs of unstable physical phenomena by linear analysis such as frequency analysis, and the development of new technology has been desired.

The present invention has been made in view of the above-mentioned problems, and an object of the present invention is to detect an abnormal state that can appropriately detect the occurrence of an abnormal state, which is an unstable physical phenomenon of a gas turbine, by a relatively simple analysis method. The point is to provide an apparatus and a method for detecting an abnormal state.

The abnormal state detecting device for achieving the above object is an abnormal state detecting device that detects the occurrence of an abnormal state including an unstable physical phenomenon of a gas turbine, and its characteristic configuration is as follows.
A pressure-related value detecting means for detecting a combustor-in-combustor pressure-related value related to a combustor-in-combustor pressure, which is the pressure in the combustor of the gas turbine, and
A network (that is, a graph) is constructed by the horizontal visibility graph method from the time series data of the pressure-related values in the combustor detected by the pressure-related value detecting means, and based on the discrete points sampled in a predetermined sampling period. With the horizontal network entropy derivation unit that derives the horizontal network entropy as the generalized network entropy (graph entropy) of the network constructed by using the horizontal visibility graph method calculated from the order distribution of the network constructed in ,
An abnormal state detecting unit for detecting whether or not the abnormal state occurs in the gas turbine based on the horizontal network entropy sequentially derived by the horizontal network entropy deriving unit is provided.
_{The horizontal network entropy is the horizontal network entropy h h, q} defined by the following [Equation 1] using the horizontal visible graph method.
The abnormal state detecting unit is at a point of detecting whether or not the abnormal state occurs in the gas turbine based on the _{horizontal network entropy h h, q.}

In the present invention, the abnormal state of the gas turbine is a concept including unstable physical phenomena such as combustion vibration, blowout, flutter, surge and turning stall.
According to the above-mentioned feature configuration, the horizontal network entropy derivation unit constructs a network by the horizontal visible graph method from the time series data of the pressure-related values in the combustor detected by the pressure-related value detecting means, and within a predetermined sampling period. _{The horizontal network entropy h h, q} calculated from the order distribution of the network constructed based on the sampled discrete points is derived, and the anomaly state detector tells the gas turbine based on the _{horizontal network entropy h h, q.} Since it is detected whether or not an abnormal state occurs, it is not necessary to execute a process such as a fast Fourier transform that increases the amount of calculation as in the prior art, and it is not necessary to provide a computer having high calculation ability. Equipment costs can be reduced.
_{Furthermore, as a result of diligent studies, the inventors have found that the horizontal network entropy h h, q} calculated from the order distribution of the horizontal visible graph constructed from the time series data of the pressure-related values in the combustor of the gas turbine is the gas turbine. It was found that the value has a high correlation with the disorder of the waveform of the pressure-related value in the combustor, which is strongly related to the abnormal state of. Based on this finding, it is possible to appropriately detect the occurrence of an abnormal state of the gas turbine by detecting whether or not an abnormal state occurs in the gas turbine based on _{the horizontal network entropy h h, q.}
In particular, the horizontal visibility graph method can be applied to phenomena where the algorithm for constructing the network is simpler and faster than other ones (for example, the normal visibility graph method), and the combustion vibration (periodic vibration) generated in the gas turbine. Since the high-speed transition process with stable combustion (aperiodic vibration) can be suitably detected, combustion vibration as an abnormal state of the gas turbine can be appropriately detected.
From the above, it is possible to realize an abnormal state detection device that can appropriately detect the occurrence of an unstable physical phenomenon of a gas turbine by a relatively simple analysis method.

Here, the visible graph is a network constructed by connecting the vertices corresponding to the discrete points with branches when the discrete points of the time series are regarded as a bar graph and visibility exists at a certain two discrete points. It is a general term for graphs. The number of branches extending from a vertex corresponding to a certain discrete point is called the degree k of the vertex. Further, there are two types of visible graphs, one is simply called a visible graph and the other is called a horizontal visible graph. The horizontal visible graph is an arbitrary time series in which the horizontal axis is time t and the vertical axis is pressure p. two discrete points _{i (t i, p i)} , j (t j, p j) all discrete points _{n (t n, p n)} that is between the when satisfies the following expression (2), discrete It is a graph constructed by determining that the points i and j are horizontally visible and connecting the vertices on the network corresponding to the discrete points i and j.

ただし、ある頂点と、その両隣の頂点の間は必ず可視である為、枝を繋ぐ。

However, since it is always visible between a certain vertex and the vertices on both sides of it, connect the branches.

For example, the description will be made based on one time series of FIG. 3 in which the horizontal axis is the time t and the vertical axis is the pressure p.
Above the bar graph of FIG. 3, the discrete point _{a (t a, p a)} and _{f (t f, p f)} all discrete points _{b (t b, p b)} between _{c (t c, p c} ) and _{d (t d, p d)} and _{e (t e, p e)} satisfies the relationship of expression (2), between the discrete points a-f is because it is visible, discrete points a, corresponding to f Connect branches between vertices. In the horizontally visible graph, the number of branches (order k) connecting the vertices corresponding to the discrete points is 3, at the apex a on the network, 2 at the apex b, and at the apex c, as shown in the lower graph of FIG. 5. Vertex d is 2, vertex e is 3, and vertex f is 3.
Incidentally, when the vertices on the network corresponding to two discrete points are connected by a straight line (not shown) between the discrete points as shown in the upper graph of FIG. 3, when the straight line intersects the bar graph, It shall not be connected by branches.

_{Since the horizontal network entropies h h and q} calculated from the order distribution of the horizontally visible graph are values having a high correlation with the disorder of the waveform of the pressure-related value in the combustor, the horizontally visible graph is as shown in the above feature configuration. _{By detecting whether or not an abnormal state occurs in the gas turbine based on the horizontal network entropy h h, q} calculated from the order distribution of, it is possible to deal with the disorder of the waveform of the pressure-related value in the combustor. It is possible to appropriately detect an abnormal state of the gas turbine (for example, blowout or combustion vibration) that occurs.

_{In particular, in the horizontal network entropy h h, q} calculated from the order distribution of the horizontal visible graph, by setting the value of q variably according to the target data, the abnormal state of the gas turbine (for example, blowout or combustion vibration) can be determined. The detection can be performed more appropriately.

Further feature configuration of the abnormal state detection device is
A storage unit is provided which derives and stores the blowout detection threshold value in advance from the _{horizontal network entropies h h and q} calculated from the order distribution in the state immediately before the blowout as the abnormal state of the gas turbine occurs.
The abnormal state detecting unit detects that the blowout occurs when the _{horizontal network entropy h h, q} derived by the horizontal network entropy derivation unit exceeds the blowout detection threshold value.

As described above, the horizontal network entropy h _{h, q} related to the horizontal visibility graph is found to be a value having a high correlation with the disorder of the waveform of the pressure-related value in the combustor, and is blown out. It was found that _{the horizontal network entropy h h and q} become higher as the possibility of occurrence of the above increases and the disturbance of the waveform of the pressure-related value in the combustor increases.
That is, as in the above feature configuration, by providing an abnormal state detection unit that detects that blowout occurs when _{the horizontal network entropy h h, q derived by the horizontal network entropy derivation unit exceeds the blowout detection threshold value.} , It is possible to appropriately detect the occurrence of blowout.

Further feature configuration of the abnormal state detection device is
Combustion vibration that is a value between the _{horizontal network entropy h h, q} when combustion vibration is generated in the gas turbine and the horizontal network entropy h _{h, q} when stable combustion is generated. It is equipped with a storage unit that derives and stores the detection threshold in advance.
The abnormal state detection unit detects that the combustion vibration is generated when the _{horizontal network entropy h h, q} derived by the horizontal network entropy derivation unit falls below the combustion vibration detection threshold value.

An abnormality that detects that combustion vibration occurs when the _{horizontal network entropy h h, q} calculated from the order distribution of the horizontal visible graph derived by the horizontal network entropy derivation unit falls below the combustion vibration detection threshold as in the above characteristic configuration. By providing a state detection unit, in a gas turbine where a high-speed transition state between combustion vibration (periodic vibration) and stable combustion (aperiodic vibration) may occur, combustion can be performed well in response to high-speed phenomena. Vibration can be detected appropriately at an early stage.

Further feature configuration of the abnormal state detection device is
The abnormal state detection unit
When detecting the occurrence of blowout as the abnormal state of the gas turbine, _{the q values of the horizontal network entropy h h and q} are set to a value larger than 0 and 1.0 or less.
When detecting the occurrence of oscillating combustion as the abnormal state of the gas turbine, _{the q value of the horizontal network entropy h h, q} is set to a value larger than 1.0.

Recently, the inventors have performed a calculation in which the value of q is variably set according to the target data _{in the horizontal network entropy h h, q} calculated from the order distribution of the horizontal visible graph. The calculation result is shown in FIG.
_{Incidentally, in the example shown in FIG. 6, the horizontal network entropy h h and q} were calculated from the waveform obtained by applying a normally distributed random number to the sine wave as shown in the pressure waveform shown in [Equation 3] below.
As a condition, the MATLAB function "randn" is used for noise, the number of samples is 6000 Hz x 60 s, the parameter q is changed between 0.001 and 10 with a step size of 0.001, and the amplitude of p is The sine wave and the noise were exchanged so as to be constant, and a function y (t) in which the time ratio of the noise increased as the time increased was defined and calculated.

From the calculation results of FIG. 6, the inventors have found that when noise is mixed in the waveform of the combustor pressure-related value as the target data, the horizontal network entropy h _{h, q} reacts and increases as the noise increases. The q value (a value greater than 0 and less than 1.0, which is relatively small) and the q value (from 1.0) in which the horizontal network entropy h _{h and q} show almost no reaction and do not change even if noise increases. It was found that there is a large value and a relatively large q value).

Furthermore, the inventors have found that the data of the pressure-related values in the combustor in the vicinity of the blowout may be buried in the noise in the absence of the dominant frequency, and therefore, the horizontal network entropy of the noise in the vicinity of the blowout h _{h When} detecting the occurrence of blowout as an abnormal state of the gas turbine in order to suppress the influence on _{， q, the q value of the horizontal network entropy h h ， q} is relatively set to a value larger than 0 and 1.0 or less. It was found that it is preferable to set it to a small value.
As a result, the influence of noise on the horizontal network entropy h _{h and q in the} vicinity of the blowout can be satisfactorily suppressed.
On the other hand, when detecting the occurrence of oscillating combustion as an abnormal state of the gas turbine _{, the q values of the horizontal network entropy h h and q} are set to a value larger than 1.0 and a relatively large value.
As a result, for example, the change in the disorder of the pressure fluctuation accompanying the change in the equivalent ratio in the stable combustion state can be grasped more prominently, and the change in the horizontal network entropy h _{h and q} becomes sensitive, so that the combustion vibration and blowing It is possible to obtain the effect that the threshold value can be set more easily in detecting an unstable physical phenomenon such as disappearance.

The abnormal state detection method for achieving the above object is an abnormal state detection method that detects in advance the occurrence of an abnormal state including an unstable combustion state of a gas turbine, and its characteristic configuration is as follows.
A pressure-related value detection step for detecting a combustor pressure-related value related to the combustor pressure, which is the pressure inside the combustor of the gas turbine,
The network is constructed by the horizontal visibility graph method from the time series data of the pressure-related values in the combustor detected in the pressure-related value detection step, and is constructed based on the discrete points sampled in a predetermined sampling period. A horizontal network entropy derivation step for deriving _{the horizontal network entropy h h, q} as a generalized network entropy of the network constructed by using the horizontal visibility graph method calculated from the order distribution of the network.
A point of executing an abnormal state detection step of detecting whether or not the abnormal state occurs in the gas turbine based on the _{horizontal network entropy h h and q} sequentially derived in the horizontal network entropy derivation step. It is in.

By executing the abnormal state detecting method, it is possible to obtain the same effect as that of the abnormal state detecting device described so far.

Schematic configuration diagram including a gas turbine model combustor of the abnormal state detection device according to the embodiment Functional block diagram of the control device Schematic diagram showing the time series of pressure fluctuations and the network structure constructed from the time series by the horizontal visible graph method. In a situation where the equivalence ratio is gradually reduced over time, (a) _{the root mean square Prms} of the root mean square Prms of the _{horizontal network entropy h h, q} calculated from the order distribution of the horizontal visibility graph and the pressure fluctuation of the combustor. a graph showing a change, (b) the square of ordinary normal pressure fluctuations of the network entropy h _v combustor as entropy of network constructed using conventional visibility graphs method is calculated from the degree distribution of the visibility graph Graph showing the change over time with the root mean square _Prms Graph diagram showing _{the root mean square Prms of the horizontal network entropy h h, q} and the pressure fluctuation of the combustor calculated from the order distribution of the horizontal visible graph when the equivalent ratio is changed. Graph diagram showing _{horizontal network entropy h h, q} when the q value is changed for data containing noise

The abnormal state detecting device 100 and the abnormal state detecting method according to the present embodiment relate to a relatively simple analysis method capable of appropriately detecting the occurrence of an abnormal state which is an unstable physical phenomenon of a gas turbine. In the embodiment, the abnormal state of the gas turbine is a concept including unstable physical phenomena such as combustion vibration, blowout, flutter, surge, and turning stall.
Hereinafter, the abnormal state detecting device 100 and the abnormal state detecting method will be described with reference to the drawings. The invention according to the present application detects the occurrence of an abnormal state of a gas turbine, but in the embodiment, a combustor simulating a partial element of a gas turbine engine combustor instead of the gas turbine. The description will be made based on a configuration example including the gas turbine model combustor 10 having the above.

The abnormal state detection device 100 has, as a main configuration, a pressure transducer 22 (combustor internal pressure) that detects the pressure (an example of a combustor internal pressure-related value) in the combustion chamber 11 (an example of a combustor) of the gas turbine model combustor 10. A network is constructed by the horizontal visible graph method from the time series data of the pressure detected by the pressure transducer 22 and the related value detecting means), and during a predetermined sampling period (for example, 100 msec), for example, 5000 Hz. Derivation _{of the horizontal network entropy h h, q} as the generalized network entropy of the network constructed using the horizontal visible graph method calculated from the order distribution of the network constructed based on the discrete points sampled in. An abnormal state is generated in the gas turbine model combustor 10 based on the horizontal network entropy derivation unit 30a (functional part of the control device 30) and the horizontal network entropy h _{h and q sequentially derived by the horizontal network entropy derivation unit 30a.} It is provided with an abnormal state detection unit 30b (functional part of the control device 30) that detects whether or not it occurs.

The gas turbine model combustor 10 is a premixed combustor that burns a mixture of fuel F and combustion air A, and has an intake unit 13 that receives the fuel F and combustion air A and the intake unit 13. It is composed of a combustion chamber 11 that burns an air-fuel mixture that has been received and swirled by a swirler (not shown), and a cooling chamber 12 that cools the exhaust from the combustion chamber 11. Incidentally, the cooling chamber 12 is provided with a cooling water jacket (not shown), and the cooling water C flows in through the cooling water inflow portion 12b and the cooling water C flows out through the cooling water outflow portion 12a. Cooling water flows through the cooling water jacket.

The intake unit 13 is communicatively connected to a fuel tank 61 for storing fuel such as methane via a fuel supply path, and the fuel supply path is a main fuel supply path L1 and a sub fuel supply path L2 in the middle flow path thereof. And are configured in parallel. Further, the main fuel supply path L1 is provided with a first mass flow controller 32 that controls the mass flow rate of the fuel F flowing through the main fuel supply path L1, and the auxiliary fuel supply path L2 is provided with the auxiliary fuel supply path. A second mass flow controller 33 that controls the mass flow rate of the fuel F passing through L2 is provided. Each of the first mass flow controller 32 and the second mass flow controller 33 is configured to be able to control the mass flow rate according to a control command from the control device 30.

Further, the intake portion 13 is communicatively connected to the compressor 62 that pumps the combustion air A via the combustion air supply path L4, and the combustion air supply path L4 is passed through the combustion air supply path L4. An air flow rate control device 51 for controlling the flow rate of the flowing combustion air A is provided, and the air flow rate control device 51 is configured to be able to control the flow rate of the combustion air A according to a control command from the control device 30. ing.

The combustion chamber 11 is provided with a pressure transducer 22 that detects the pressure at the wall surface position near the inlet of the combustion chamber 11 and outputs a voltage signal corresponding to the detected pressure. Further, an amplifier 21 for amplifying the voltage signal output from the pressure transducer 22 is provided, and the voltage signal amplified by the amplifier 21 is transmitted to the control device 30.
That is, the pressure transducer 22 and the amplifier 21 function as pressure-related value detecting means.

As shown in the functional block diagram of FIG. 2, the control device 30 has a horizontal network entropy derivation unit 30a as a functional part for detecting whether or not an abnormal state of the gas turbine model combustor 10 described above occurs, and an abnormality. It has a state detection unit 30b, a storage unit 30c, and a threshold value setting unit 30d. Each functional part is implemented in a form in which hardware including a CPU, a memory, an input / output device, and the like and software including a computer program group cooperate with each other.

The voltage signal corresponding to the pressure transmitted from the amplifier 21 is input to the horizontal network entropy deriving unit 30a of the control device 30 via the pressure signal input unit 31 as an input / output board.
Incidentally, the horizontal network entropy derivation unit 30a stores a voltage signal corresponding to the pressure for a predetermined period (for example, 100 msec) in a memory (not shown) in a FIFO format, and based on the stored voltage signal, the horizontal network entropy is as follows. h _{h, q} are derived.

The horizontal network entropy derivation unit 30a constructs a horizontal visible graph from a voltage signal (time series data of pressure) corresponding to pressure, and at a predetermined sampling period (for example, 100 msec), for example, at a discrete point sampled at 5000 Hz. _{The horizontal network entropies h h and q} are derived from the order distribution of the network constructed based on this.
In the embodiment, the horizontal network entropy derivation unit 30a derives the horizontal network entropy h _{h, q} defined in the following [Equation 1] by using the horizontal visibility graph method.

Here, the visible graph is a network in which discrete points in a time series are regarded as a bar graph, and when visibility exists at two discrete points, the vertices on the network corresponding to the discrete points are connected by branches, or It is a general term for graphs. Here, the number of branches is referred to as the degree k of the vertices on the network corresponding to the discrete points. Furthermore, the horizontal visibility graphs in visibility graph, the horizontal axis indicates time t, the vertical axis is described as an example a time series of the pressure p, any two discrete points _{i (t i, p i)} , j ( When all the discrete points n (t _n , p _{n ) between t j} , p _j ) satisfy the following [Equation 2], the vertices on the network corresponding to the discrete points i, j are connected. It is a graph constructed by.

ただし、ある頂点と、その両隣の頂点の間は必ず可視である為、枝を繋ぐ。

However, since it is always visible between a certain vertex and the vertices on both sides of it, connect the branches.

For example, the description will be given based on the time series of FIG. 3 in which the horizontal axis is the time t and the vertical axis is the pressure p.
Above the bar graph of FIG. 3, the discrete point _{a (t a, p a)} and _{f (t f, p f)} all discrete points _{b (t b, p b)} between _{c (t c, p c} ) and _{d (t d, p d)} and _{e (t e, p e)} satisfies the relationship of expression (2), since the inter discrete points a-f are visible, corresponding to between discrete points a-f Connect branches to the vertices on the network.
In the horizontally visible graph, the number of branches (order k) connecting the vertices on the network corresponding to the discrete points is 3 at the apex a, 2 at the apex b, and 2 at the apex c, as shown in the lower graph of FIG. 5. Vertex d is 2, vertex e is 3, and vertex f is 3.
By the way, the vertices on the network corresponding to the two discrete points are not connected by branches when the straight lines intersect with the bar graph when the discrete points are connected by a straight line as shown in the upper graph of FIG. And.

The threshold value setting unit 30d sets a threshold value for determining an abnormal state of the gas turbine model combustor 10. The threshold value setting unit 30d may receive information indicating the threshold value from the user, or may receive information indicating the threshold value given by other software. The received threshold value is stored in the storage unit 30c.
In the embodiment, the storage unit 30c stores as a threshold value the blowout detection threshold value T1 which is _{the horizontal network entropy h h, q} in the state immediately before the blowout as an abnormal state of the gas turbine model combustor 10. doing. _{Further, the storage unit 30c has horizontal network entropies h h and q} when combustion vibration (periodic vibration) is generated in the gas turbine model combustor 10 and stable combustion (periodic vibration) when stable combustion (periodic vibration) is generated. The combustion vibration detection threshold T2, which is a value between the horizontal network entropies h _{h and q of, is stored.}

In the embodiment, the abnormal state detection unit 30b detects that the gas turbine model combustor 10 is blown out as an abnormal state and that combustion vibration is generated.
_{In the abnormal state detection unit 30b, whether or not the horizontal network entropy h h and q} sequentially derived by the horizontal network entropy derivation unit 30a exceeds the above-mentioned blowout detection threshold value T1 and whether or not it is below the above-mentioned combustion vibration detection threshold value T2. Monitor whether or not.
_{When the horizontal network entropy h h, q} sequentially derived by the horizontal network entropy derivation unit 30a exceeds the blowout detection threshold value T1, the abnormal state detection unit 30b detects that there is a high possibility that blowout will occur thereafter, and is horizontal. _{When the horizontal network entropies h h and q} sequentially derived by the network entropy derivation unit 30a are below the combustion vibration detection threshold value T2, it is detected that there is a high possibility that combustion vibration will occur thereafter.

Although detailed description will be omitted, in the embodiment, the control device 30 controls the normal output of the gas turbine model combustor 10 by using the first mass flow controller 32 to control the mass flow rate of the fuel F and the air flow rate control device. It is executed by controlling the flow rate of the combustion air A by 51, but when the abnormal state detection unit 30b detects that there is a high possibility that the blowout will occur, the mass of the fuel F is measured by the second mass flow controller 33. Control to gradually increase the flow rate is executed, and when the abnormal state detection unit 30b detects that there is a high possibility that combustion vibration will occur, the second mass flow controller 33 executes control to reduce the mass flow rate of the fuel F. do.

That is, in the abnormal state detection method of the abnormal state detection device 100 according to the embodiment, the pressure-related value detecting step for detecting the pressure of the gas turbine model combustor 10 (an example of the pressure-related value in the combustor) and the pressure-related value. A network is constructed by the horizontal visible graph method from the time-series data of the pressure detected in the detection step, and the network is constructed based on the discrete points sampled at, for example, 5000 Hz in a predetermined sampling period (for example, 100 msec). The horizontal network entropy derivation step of deriving the horizontal network entropy _{h h, q} (an example of the order distribution) defined in the above [Equation 1] using the order distribution of, and the horizontal network entropy derivation step are sequentially derived. Based on the horizontal network entropies h _{h and q} , the abnormal state detection step of detecting whether or not an abnormal state occurs in the gas turbine model combustor 10 is executed.

Next, the test results for verifying whether or not the abnormal state of the gas turbine model combustor 10 can be detected by the abnormal state detecting device 100 and the abnormal state detecting method according to the above embodiment are shown with reference to FIGS. explain. However, in the following, the case where q = 1 is set will be described as an example.

Incidentally, in the test results, for comparison, also shows typical test results usually based on network entropy h _v using degree distribution of the visibility graph.
Incidentally, a normal visible graph, the horizontal axis represents time t, when the time series of the vertical axis the pressure p as an example, any two discrete points _{i (t i, p i)} , j (t j, p j) A graph constructed by connecting vertices on the network corresponding to discrete points i and j when all the discrete points n (t _n , p _{n) between are satisfied with the following [Equation 4].} Is.

ただし、ある頂点と、その両隣の頂点の間は必ず可視である為、枝を繋ぐ。

However, since it is always visible between a certain vertex and the vertices on both sides of it, connect the branches.

Furthermore, the test result of the normal network entropy h _v using degree distribution of the normal visibility graphs described below introduces a vision N _vis that limits the range that can be visualized. N _vis has the effect of reducing the effects of outliers and reducing the increase in order due to low frequencies. That is, the increase in the order is suppressed by applying _Nvis when the data in which the dominant frequency does not exist is buried in the noise, such as in the vicinity of the blowout. Since there is no predominant frequency in the pressure fluctuation in the vicinity of the blowout, the predominant frequency of the combustion vibration is set to the frequency band f to be detected, and the predominant frequency of the combustion vibration fluctuates in about ± 5 to 10 Hz. 5] determines _{N vis.} However, f _s is used as the sampling frequency.

Using the visual acuity N _{vis , the normal network entropy h v} using the order distribution of the normal visible graph is expressed based on the following [Equation 6].

In the test, the equality ratio φ of the air-fuel mixture introduced into the gas turbine model combustor 10 is kept constant at 0.84 at 0 ≦ t [sec] ≦ 10, and is 0 until about 10 <t [sec] ≦ 40. root mean square P _rms of the time-series data of the pressure when gradually reduced at a constant rate from .84 to _0.50, the horizontal network entropy using degree distribution of horizontal visibility graph h _{h, q,} normal visible _{The normal network entropy hv} was measured using the order distribution of the graph.

Although not shown, the disorder of the pressure waveform is low in the combustion vibration range (periodic vibration range) of 0 ≦ t [sec] ≦ 10 in FIG. 4, and stable combustion of about 10 <t [sec] ≦ 40. It is known to be high in the region (aperiodic vibration region). Further, the disorder of the pressure waveform may increase in the stable combustion region (aperiodic vibration region) of about 10 <t [sec] ≦ 40 as the equivalent ratio φ decreases and approaches the timing of extinction. Are known. The randomness of the pressure waveform is a value that is considered to have a high correlation with abnormal conditions such as blowout, combustion vibration, surge, flutter, and turning stall of the gas turbine model combustor 10, and is highly correlated with the value. It is important to discover the parameter having the above in detecting the abnormal state of the gas turbine model combustor 10.
The "randomness of the pressure waveform" means "the disorder of the network structure constructed from the pressure waveform". If the network is constructed with various orders, the network structure is not biased and the horizontal network entropy h _{h, q} increases. From the viewpoint of information theory, it can be judged that the network structure becomes more disordered as the _{horizontal network entropies h h and q increase.}

As shown in FIG. 4 (a), P _rms is the 10 <t (sec) ≦ 40 about stable combustion zone, corresponding to the decrease of equivalence ratio φ over time, randomness of the pressure waveform is high Nevertheless, it shows a substantially constant value. Therefore, P _rms can not be said to have a high correlation with the randomness of the pressure waveform. Therefore, even if a threshold value or the like is provided, it can be said that it is difficult to detect the occurrence of an abnormal state such as blowout of the gas turbine model combustor 10 by using _Prms.
On the other hand, h _{h and q} are low in the combustion vibration range (periodic vibration range) of 0 ≦ t [sec] ≦ 10 as shown in FIG. 4 (a), and are about 10 <t [sec] ≦ 40. In the stable combustion region (aperiodic vibration region), there is a tendency for the equivalence ratio φ to gradually increase with the passage of time at a substantially constant rate, which is highly correlated with the disorder of the pressure waveform. You can see that it has.
In particular, h _{h and q} gradually increase at a substantially constant rate as the equivalent ratio φ decreases with the passage of time in the stable combustion region (aperiodic vibration region) of about 10 <t [sec] ≦ 40. Since there is a tendency, for example, as shown in FIG. 4A, by setting the blowout detection threshold T1 and the combustion vibration detection threshold T2, the blowout and combustion vibration of the gas turbine model combustor 10 can be set. It is possible to detect the occurrence of an abnormal state.
That is, as a result, it is possible to detect a sign of an abnormal state such as blowout or combustion vibration.

On the other hand, _{as shown in FIG. 4B, the normal network entropy hv} is high in the combustion vibration range (periodic vibration range) of 0 ≦ t [sec] ≦ 10, and is stable with about 10 <t [sec] ≦ 40. In the combustion region (aperiodic vibration region), there is a tendency for the equivalence ratio φ to gradually decrease with the passage of time at a substantially constant rate, and the network entropy _hv is usually the disorder of the pressure waveform. Cannot be handled properly.
Therefore, it cannot be said that it is suitable for detecting various abnormal states of the gas turbine model combustor 10.

Incidentally, the graph of FIG. 5 _{shows h, q} and _Prms for each equivalent ratio without time, but also in the graph of FIG. 5, the equivalent ratio φ is h. _{It can be seen that h, q} and _Prms have the same tendency as those described in FIG. 4 above.

Since the q values of the horizontal network entropy h _{h and q} can be set to various values larger than 0, the inventors have now set the horizontal network entropy h _{h and q} calculated from the order distribution of the horizontal visible graph. , A calculation was performed to set the value of q variably according to the target data. The calculation result is shown in FIG.
_{Incidentally, in the example shown in FIG. 6, the horizontal network entropy h h and q} were calculated from the waveform obtained by applying a normally distributed random number to the sine wave as shown in the pressure waveform shown in [Equation 3] below.
As a condition, the MATLAB function "randn" is used for noise, the number of samples is 6000 Hz x 60 s, the parameter q is changed between 0.001 and 10 with a step size of 0.001, and the amplitude of p is The sine wave and the noise were exchanged so as to be constant, and a function y (t) in which the time ratio of the noise increased as the time increased was defined and calculated.

From the calculation results of FIG. 6, the inventors have found that when noise is mixed in the waveform of the combustor pressure-related value as the target data, the horizontal network entropy h _{h, q} reacts and increases as the noise increases. The q value (a value greater than 0 and less than 1.0, which is relatively small) and the q value (from 1.0) in which the horizontal network entropy h _{h and q} show almost no reaction and do not change even if noise increases. It was found that there is a large value and a relatively large q value).

Furthermore, the inventors have found that the data of the pressure-related values in the combustor in the vicinity of the blowout may be buried in the noise because the data in which the dominant frequency does not exist may be buried in the noise. _When detecting the occurrence of blowout as an abnormal state of the gas turbine in order to suppress the influence on _{h and q} , the q values of the horizontal network entropy h h and q are compared with values larger than 0 and 1.0 or less. It was found that it is preferable to set the value to a small value.
As a result, the influence of noise on the horizontal network entropy h _{h and q in the} vicinity of the blowout can be satisfactorily suppressed.
On the other hand, when detecting the occurrence of oscillating combustion as an abnormal state of the gas turbine _{, the q values of the horizontal network entropy h h and q} are set to a value larger than 1.0 and a relatively large value.
This has the effect of making it easier to set the threshold value in detecting unstable physical phenomena such as combustion vibration and blowout.

[Another Embodiment]
(1) In the above embodiment, an example is shown in which the pressure-related value in the combustor is the pressure in the combustor of the gas turbine. However, the pressure-related value in the combustor may be the self-luminous intensity of ^{OH *.}

(2) In the above-described embodiment, the configuration for monitoring the combustion state in the gas turbine model combustor 10 has been described, but the present invention is not limited to this. For example, a combustor mounted on a jet engine or an industrial gas turbine engine may be targeted.

(3) In the above embodiment, an example in which the fuel flow rate control is executed based on the detection of the abnormal state detection unit 30b is shown, but the fuel flow rate control may not be executed separately.

(4) A voltage signal corresponding to a pressure of 100 msec is stored in a memory for use in the calculation of the horizontal network entropy derivation unit 30a, and while updating the voltage signal of the memory, the horizontal network entropy h _{h is repeated every 100 msec. The configuration for deriving ， q} has been described, but the present invention is not limited to this. For example, it may be configured to calculate based on a value in a period other than 100 msec, such as 50 msec.

(5) Further, in the above-described embodiment, the control device 30 is realized by a computer, but the present invention is not limited to this. For example, hardware capable of executing combustion control processing may be configured by ASIC, FPGA, or the like, and this may be used as a control device. Further, instead of configuring the control device 30 by one computer, it may be a distributed system in which the above-mentioned arithmetic processing is executed by distributed processing by a plurality of computers.

The configuration disclosed in the above embodiment (including another embodiment, the same shall apply hereinafter) can be applied in combination with the configuration disclosed in other embodiments as long as there is no contradiction. The embodiments disclosed in the present specification are examples, and the embodiments of the present invention are not limited thereto, and can be appropriately modified without departing from the object of the present invention.

The abnormal state detecting device and the abnormal state detecting method of the present invention are an abnormal state detecting device and an abnormal state that can appropriately detect the occurrence of an abnormal state, which is an unstable physical phenomenon of a gas turbine, by a relatively simple analysis method. It can be effectively used as a detection method.

10: Gas turbine model combustor 21: Amplifier 22: Pressure transducer 30: Control device 30a: Horizontal network entropy derivation unit 30b: Abnormal state detection unit 30c: Storage unit 100: Abnormal state detection device T1: Blowout detection threshold T2: Combustion Vibration detection threshold h _{h, q} : Generalized network entropy of the network constructed using the horizontal visibility graph method h _v : Network entropy of the network constructed using the ordinary visibility graph method k: Order p: Pressure

Claims (5)

An abnormal condition detection device that detects the occurrence of abnormal conditions including unstable physical phenomena of gas turbines.
A pressure-related value detecting means for detecting a combustor-in-combustor pressure-related value related to a combustor-in-combustor pressure, which is the pressure in the combustor of the gas turbine, and
The network is constructed by the horizontal visibility graph method from the time series data of the pressure-related values in the combustor detected by the pressure-related value detecting means, and is constructed based on the discrete points sampled in a predetermined sampling period. A horizontal network entropy derivation unit that derives the horizontal network entropy as a generalized network entropy of the network constructed by using the horizontal visibility graph method calculated from the order distribution of the network, and a horizontal network entropy derivation unit.
An abnormal state detecting unit for detecting whether or not the abnormal state occurs in the gas turbine based on the horizontal network entropy sequentially derived by the horizontal network entropy deriving unit is provided.
_{The horizontal network entropy is the horizontal network entropy h h, q} defined by the following [Equation 1] using the horizontal visible graph method.
The abnormal state detecting unit is an abnormal state detecting device that detects whether or not the abnormal state occurs in the gas turbine based on the _{horizontal network entropy h h, q.}

A storage unit is provided which derives and stores a blowout detection threshold value in advance from the horizontal network entropy calculated from the order distribution in the state immediately before the blowout as the abnormal state of the gas turbine occurs.
The first aspect of claim 1, wherein the abnormal state detecting unit detects that the blowout occurs when the _{horizontal network entropy h h, q derived by the horizontal network entropy derivation unit exceeds the blowout detection threshold value.} Abnormal state detection device. Combustion vibration that is a value between the _{horizontal network entropy h h, q} when combustion vibration is generated in the gas turbine and the horizontal network entropy h _{h, q} when stable combustion is generated. It is equipped with a storage unit that derives and stores the detection threshold in advance.
According to claim 1 or 2, the abnormal state detection unit detects that the combustion vibration is generated when the _{horizontal network entropy h h, q} derived by the horizontal network entropy derivation unit falls below the combustion vibration detection threshold value. The described abnormal condition detection device. The abnormal state detection unit
When detecting the occurrence of blowout as the abnormal state of the gas turbine, _{the q values of the horizontal network entropy h h and q} are set to a value larger than 0 and 1.0 or less.
Any one of claims 1 to 3 in which the q values of the _{horizontal network entropy h h and q} are set to a value greater than 1.0 when detecting the occurrence of oscillating combustion as the abnormal state of the gas turbine. Abnormal state detection device described in. It is an abnormal state detection method that detects the occurrence of abnormal states including unstable physical phenomena of gas turbines in advance.
A pressure-related value detection step for detecting a combustor pressure-related value related to the combustor pressure, which is the pressure inside the combustor of the gas turbine,
The network is constructed by the horizontal visibility graph method from the time series data of the pressure-related values in the combustor detected in the pressure-related value detection step, and is constructed based on the discrete points sampled in a predetermined sampling period. A horizontal network entropy derivation process for deriving the horizontal network entropy as a generalized network entropy of the network constructed by using the horizontal visibility graph method calculated from the order distribution of the network, and a process for deriving the horizontal network entropy.
Based on the horizontal network entropy sequentially derived in the horizontal network entropy derivation step, an abnormal state detection step of detecting whether or not the abnormal state occurs in the gas turbine is executed.
_{The horizontal network entropy is the horizontal network entropy h h, q} defined by the following [Equation 1] using the horizontal visible graph method.
The abnormal state detection step is an abnormal state detection method for detecting whether or not the abnormal state occurs in the gas turbine based on the _{horizontal network entropy h h, q.}

Images