JP2016173211A - Observation system and observation method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To provide an observation system and an observation method for physical phenomenon.SOLUTION: This invention comprises a detecting part for detecting physical quantity of physical phenomenon that varies in view of time-variant, a generating part 313 for setting each physical quantity at each time as a node and generating a complex network connecting the nodes by an edge in view of time-variant in reference to conditions [A] and [B] indicated below, an operation part 314 for calculating a prescribed feature amount in the complex network, an estimation part 316 for estimating a state of physical phenomenon on the basis of the feature amount. [A]: nodes that are set by the physical quantity of which detection times are adjacent to each other; [B] nodes that are set by the physical quantity yand the physical quantity yin the case that the physical quantity ydetected at the time t, the physical quantity ydetected at the time tlater than the time tand physical quantity ydetected at the time tbetween the time tand the time tsatisfy the following formula.SELECTED DRAWING: Figure 2

Description

  The present invention relates to an observation apparatus and an observation method for observing a physical phenomenon.

  For example, in a gas turbine combustor, the occurrence of unstable combustion conditions such as combustion vibrations and blow-offs is undesirable. Further, in the gas turbine engine, in addition to such an unstable combustion state, unstable phenomena such as rotor blade flutter, turning stall, and surge may occur. Also, in fields other than gas turbine engines, undesired unstable fluid flow conditions may occur, for example, in heat exchangers or chemical reactors. Conventionally, it has been desired to detect the occurrence of such an unstable physical phenomenon.

  Some unstable physical phenomena occur suddenly, such as blowouts in a gas turbine combustor. Such an unexpectedly unstable physical phenomenon is difficult to detect accurately by linear analysis such as frequency analysis. Therefore, conventionally, a method for detecting an unstable physical phenomenon using nonlinear analysis has been developed. For example, Patent Document 1 discloses a method for measuring the fluctuation of a flame by adopting a method used in chaos time series analysis. In the method described in Patent Document 1, the attractor is reconfigured using the Turkens embedding theorem, the correlation dimension and the maximum Lyapunov exponent are obtained, and the combustion state is measured from these.

  However, the correlation dimension and the Lyapunov exponent used in the method described in Patent Document 1 are classical nonlinear analysis theories, and in particular, the correlation dimension requires time-series data with an enormous number of data points. Therefore, the calculation load is high, and the calculation process can be performed only after acquiring a collection of time series data. For this reason, the method described in Patent Document 1 is not suitable for an application in which occurrence of a physical phenomenon is estimated with high immediacy, and its application field is limited.

Patent Document 2 discloses a method of detecting a physical quantity that changes over time, calculating a translation error related to the detected physical quantity, and estimating the occurrence of a specific physical phenomenon based on the calculated translation error. Has been.
The method described in Patent Document 2 can reduce the number of data of physical quantities used for calculation compared to the method described in Patent Document 1, and can greatly reduce the amount of data and increase the occurrence of a specific physical phenomenon. Can be estimated by immediacy. Then, by using the method described in Patent Document 2, a change in pressure accompanying combustion in a gas turbine combustor or the like is detected, and the occurrence of an unstable combustion state is estimated from this pressure change, thereby providing a fuel supply amount. To control.

JP 2006-138517 A JP 2013-238365 A

  However, when it is necessary to estimate physical phenomena in a short time with high immediacy, such as control of fuel supply according to unstable combustion conditions in the combustor, a simpler algorithm is used. There is a need for processing with higher responsiveness and stability in order to increase the operation speed.

  An object of the present invention is to provide an observation apparatus and an observation method that enable observation of a physical phenomenon having higher responsiveness and the like by a method that is different from the conventional one.

(1) The observation apparatus according to the present invention is:
A detection unit for detecting a physical quantity of a physical phenomenon that changes over time;
A generation unit that sets a physical quantity detected at each time as a node, and generates a complex network in which a plurality of nodes are connected by an edge according to the following conditions (A) and (B):
A calculation unit for obtaining a predetermined feature amount in the complex network;
An estimation unit that estimates a state of a physical phenomenon based on the feature amount.
(A): a node between which is set by the physical quantity detection time is adjacent (B): and the time t _a detected physical quantity y _a, a physical quantity y _b detected at time t _b after time t _a , the node between the physical quantity y _c detected at time t _c between time t _a and the time t _b is in the case which satisfies the following equation, is set by the physical quantity y _a physical quantity y _b

According to this observation apparatus, a complex network for a physical quantity of a physical phenomenon is generated according to a predetermined condition, and the state of the physical phenomenon is estimated based on the feature quantity of the complex network. As a result, the state of the physical phenomenon can be estimated by a simple algorithm, the calculation can be speeded up, and processing with higher response can be realized.
As the feature amount of the complex network, an average order, an average vertex distance, a cluster coefficient, a mode value of the order distribution, a frequency of the mode value of the order distribution, and the like can be used. The generation unit may generate an adjacency matrix or an adjacency list for describing a complex network, but is not limited to this, and data (number of nodes) necessary for simply calculating a predetermined feature amount is not limited thereto. , The number of edges, etc.) may be acquired. In this case, the processing as the generation unit may be substantially included in the processing as the calculation unit.

(2) It is preferable that the generation unit restricts a range of nodes connected by an edge based on a predetermined rule in the complex network.
With such a configuration, the amount of calculation can be reduced efficiently, and the calculation speed can be further increased.

(3) The range of the node is preferably set based on the number of times the physical quantity is detected per change period of the physical quantity when a predetermined physical phenomenon occurs.
With such a configuration, the number of edges connected to a node when a predetermined physical phenomenon occurs, that is, an unintentional increase in the degree can be suppressed, and occurs when a predetermined physical phenomenon occurs. It is possible to make a difference in order between when there is no time. Since the feature amount of a complex network usually has a high relationship with the order, by adopting the above configuration, it can be suitably used for estimating the state of a physical phenomenon.

(4) The feature amount may be an average order of the complex network.
With such a configuration, the network feature amount can be obtained by a very simple calculation.

(5) The estimation unit may estimate a state of a physical phenomenon based on a comparison between the average order and a predetermined threshold.

(6) The physical phenomenon may be combustion of fuel in a combustor, and the physical quantity may be a pressure in the combustor.
In this case, the state of the physical phenomenon can be estimated from the change in pressure accompanying combustion.

(7) In the above (6), it is preferable that the estimation unit estimates instability of combustion in the combustor.
With such a configuration, it is possible to estimate the occurrence of combustion instability such as combustion vibration or blow-off, and use it for fuel supply control or the like.

(8) It is preferable that the estimation unit estimates the occurrence of blow-out as unstable combustion in the combustor.
In recent years, in a combustor, the generation of carbon dioxide and NOx has been reduced by diluting the premixed gas, but this dilution increases the possibility that blowout will occur. It is extremely useful to estimate the occurrence of blow-out as in the invention.

(9) The observation method according to the present invention is:
Detecting a physical quantity of a physical phenomenon that changes over time;
A step of setting a detected physical quantity at each time as a node and generating a complex network in which a plurality of nodes are connected by an edge according to the following conditions (A) and (B):
Obtaining a predetermined feature amount in the complex network;
Estimating a state of a physical phenomenon based on the feature amount.
(A): a node between which is set by the physical quantity detection time is adjacent (B): and the time t _a detected physical quantity y _a, a physical quantity y _b detected at time t _b after time t _a , the node between the physical quantity y _c detected at time t _c between time t _a and the time t _b is in the case which satisfies the following equation, is set by the physical quantity y _a physical quantity y _b

  According to this observation method, a complex network for a physical quantity of a physical phenomenon is generated according to a predetermined condition, and the state of the physical phenomenon is estimated based on the feature quantity of the complex network. As a result, the state of the physical phenomenon can be estimated by a simple algorithm, the calculation can be speeded up, and processing with higher response can be realized.

  According to the present invention, a highly responsive physical phenomenon can be observed.

It is a schematic diagram which shows the structure of the observation apparatus of the combustion state which concerns on embodiment. It is a functional block of a signal analyzer. It is a graph which shows an example of the pressure fluctuation signal used for generation of a complicated network. It is explanatory drawing which shows an example of a general complicated network (graph; Graph). It is explanatory drawing which shows an example of a visible graph. (A) is a figure which shows the result of having connected the vertex (node) with the branch (edge) in the visible graph, and (b) is a figure which shows the complex network produced | generated from the result. It is a figure which shows an example of the range (sight) of the node connected with an edge on a visible graph. It is a figure which shows the other example of the range (sight) of the node connected with an edge on a visible graph. It is a graph for demonstrating the setting of a visual acuity. It is a graph which shows the relationship between the average order of a complex network, and an equivalence ratio. It is a flowchart which shows the procedure of a combustion control process. It is a flowchart which shows the procedure of a secondary fuel flow rate calculation process. It is a graph which shows the result of the performance evaluation test of an observation apparatus. It is a graph which shows the result of the performance evaluation test of an observation apparatus.

Hereinafter, preferred embodiments of the present invention will be described with reference to the accompanying drawings.
Below, the structure and operation | movement of an observation apparatus which observes the combustion state of a gas turbine combustor are demonstrated.

[1. Configuration of observation equipment]
The observation apparatus according to the present embodiment analyzes a combustion state (physical phenomenon) of a gas turbine model combustor (a combustor simulating a partial element of a gas turbine engine combustor) using a characteristic amount of a complex network, and calculates a combustion state. Is estimated. In particular, in the present embodiment, it is estimated that the stable vibration state and the unstable combustion state can be distinguished from the occurrence of combustion vibration and blow-off. First, the configuration of the observation apparatus will be described.

  FIG. 1 is a schematic diagram illustrating a configuration of a combustion state observation apparatus according to the present embodiment. As shown in FIG. 1, the observation apparatus 100 includes a gas turbine model combustor (hereinafter also simply referred to as “combustor”) 1, a pressure transducer 2, a control unit 3, a fuel tank 4, a compressor 5, It has.

The combustor 1 is a premixed combustor (lean premixed combustor) that burns a fuel / air mixture. The combustor 1 includes a combustion chamber 11, a water cooling chamber 12, an intake portion 13, and a swirler (not shown). The combustor 1 mixes the air supplied from the compressor 5 and the methane supplied from the fuel tank 4 at the intake portion 13 to form a premixed gas, and swirls the premixed gas with a swirler. The combustion chamber 11 is configured to burn.
The combustor 1 may be a jet engine or an industrial gas turbine engine combustor.

  A pressure transducer (detection unit) 2 is attached to the combustion chamber 11. The pressure transducer 2 detects pressure fluctuation (physical quantity) at the wall surface position of the combustion chamber 11. The pressure transducer 2 outputs a pressure fluctuation signal that is a voltage signal corresponding to the detected pressure fluctuation.

  An amplifier 21 is connected to the pressure transducer 2, and the pressure fluctuation signal output from the pressure transducer 2 is amplified by the amplifier 21.

The control unit 3 includes a signal analysis device 31 and mass flow controllers 32 and 33. The signal analysis device 31 is configured by installing in a computer a computer program that executes a combustion control process described below, and various functions in the signal analysis device 31 are realized by this computer program.
The computer constituting the signal analysis device 31 includes an arithmetic processing device, a storage unit, an input / output device, and the like.

  FIG. 2 shows functional blocks of the signal analysis device 31. This signal analysis device 31 includes a pressure fluctuation signal input unit 311, a storage unit 312, a complex network generation unit 313, an average order calculation unit 314, a threshold setting unit 315, a blowout occurrence estimation unit 316, a secondary fuel flow rate calculation unit 317, And a function as a secondary fuel flow rate control unit 318.

  The pressure fluctuation signal input unit 311 is for receiving the pressure fluctuation signal output from the pressure transducer 2 and amplified by the amplifier 21. FIG. 3 is a graph showing changes in the pressure fluctuation signal over time.

  The storage unit 312 is for storing a pressure fluctuation signal for a predetermined period (for example, 50 msec). The pressure fluctuation in the combustion chamber 11 is detected by the pressure transducer 2 and output as a time-series pressure fluctuation signal. The pressure fluctuation signal taken in by the pressure fluctuation signal input unit 311 is stored in the storage unit 312 in the FIFO format. That is, when a new pressure fluctuation signal is added to the storage unit 312, the old pressure fluctuation signal (the pressure fluctuation signal stored about 50 msec before) is deleted. In this manner, the pressure fluctuation signal in the storage unit 312 is updated, so that the pressure fluctuation signal for a period of 50 msec is always stored.

The complex network generation unit 313 is for generating a complex network based on the pressure fluctuation signal stored in the storage unit 312.
The average order calculation unit 314 is for calculating an average order that is a feature amount of the complex network by calculation.

  Here, the complex network is a generic name for networks in which a large number of components are connected in a complicated manner. FIG. 4 illustrates a graph as an aspect of a complex network. A graph is a mathematical expression method of a network, and is composed of a plurality of nodes and edges that connect the nodes. The number of edges coming out from each node is referred to as “order”. Each node shown in FIG. 4 describes the degree (number of edges). In one network represented by a graph, the average of the number of edges coming out from each node is referred to as “average order”. Therefore, the average order <k> is expressed by the following formula (1).

Here, n is the number of nodes, and k _i is the order of each node.

For example, in the example shown in FIG. 4, the average order <k> is
<K> = (1 + 4 + 2 + 3 + 2) /5=2.4
It becomes.

  The complex network generation unit 313 of the present embodiment generates a complex network from the pressure fluctuation signal as shown in FIG. In order to generate a complex network, a “visible graph” as described below is used.

  FIG. 5 is an explanatory diagram illustrating an example of a visible graph. The visible graph connects nodes that appear to avoid obstacles (nodes with line of sight) with edges. The visible graph illustrated in FIG. 5 is configured by a bar graph, and the vertex of each bar graph is set as a node. Each node is connected by an edge to a node on the right side of itself.

  For example, when viewing the nodes n2 to n6 of the right bar graph from the node n1 of the leftmost bar graph, the nodes n2 to n5 can be seen without being blocked, but the nodes n6 and beyond (right side) can be seen. All nodes are obstructed by the bar graph having node n5 and cannot be seen. Therefore, in this case, the node n1 is connected to the nodes n2 to n5 by the edge, and is not connected to the node n6 and subsequent nodes. Thus, a complex network can be generated by connecting a plurality of nodes with edges using a visible graph.

  In the visible graph, nodes adjacent in the horizontal axis direction are always connected by edges. Whether or not the nodes one or more away in the horizontal axis direction are connected by an edge is determined by whether or not the following equation (2) is satisfied.

Here, t represents the value on the horizontal axis of the visible graph in FIG. 5, and y represents the value on the vertical axis. The subscript a indicates a node as a connection source, the subscript b indicates a node as a connection destination, and the subscript c indicates an intermediate node between the connection source and the connection destination.
Expression (2) is provided on condition that the slope of the line connecting the connection source node and the connection destination node is larger than the slope of the line connecting the connection source node and the intermediate node. When this condition is satisfied, the connection source node and the connection destination node are connected. In FIG. 5, an upward arrow indicates that the inclination is large, a downward arrow indicates that the inclination is small, and a horizontal arrow indicates that the inclination is the same.

  On the lower side of the visible graph in FIG. 5, the connection relationship when the node of the leftmost bar graph is the connection source a and the other node is the connection destination b is shown in a table. Intermediate nodes are denoted by c1, c2,. At the left end of this table, “◯” and “×” indicating whether or not connection is possible are shown. That is, if the slope of the line connecting the connection source node and the connection destination node is larger than the slope of the line connecting the connection source node and the intermediate node, both nodes are connected by an edge, so When the slope of the line connecting the connection source node and the connection destination node is smaller than the slope of the line connecting the connection source node and the intermediate node, the nodes are not connected by the edge. Therefore, “×” is attached.

  FIG. 6A is a diagram in which each node is connected by an edge according to the determination of Expression (2). As a result, the generated complex network is shown in FIG. In FIG. 6B, a horizontally extending line means an edge connecting adjacent nodes, and a line curved up and down from a horizontally extending line means an edge connecting one or more nodes apart from each other. To do.

  With the above procedure, a complex network can be generated using a visible graph. And the complex network production | generation part 313 in this embodiment produces | generates a complex network in the procedure similar to the above by making the graph of the pressure fluctuation signal shown in FIG. 3 into a visible graph. The pressure fluctuation signal shown in FIG. 3 is time-series data, and the complex network generation unit 313 can generate a complex network over time each time one or more pressure fluctuation signals are acquired. In general, in a complex network, the connection relationship of a plurality of nodes is described by an adjacency matrix or an adjacency list, but the complex network generation unit 313 of this embodiment does not generate an adjacency matrix or an adjacency list in a clear manner. In addition, the number of nodes and order data used in the average order calculation unit 314 described below are acquired. Therefore, high-speed arithmetic processing can be performed.

  The average order calculation unit 314 shown in FIG. 2 calculates the average order <k> according to the above-described equation (1) for the complex network of pressure fluctuation signals generated using the equation (2). In this embodiment, the calculation of the average order is repeated every 50 msec, and observation is performed with high immediacy.

When generating a complex network using a visible graph, for example, when a suddenly high value occurs in the time series data or the fluctuation becomes very gradual, the number of edges becomes enormous and the average order <k> also increases accordingly.
For example, as shown in FIG. 7A, when only one high value such as an outlier exists in time-series data, the prospect between that node and other nodes is improved, so the whole The order of becomes significantly higher.
Further, as shown in FIG. 8A, when the time-series data gently changes in a concave shape with a low frequency period, the visibility of the nodes is improved, and the overall order is significantly increased.

  In the present embodiment, the concept of “sight” that restricts the visibility from each node so as to suppress the order from becoming significantly high in any of the above cases and to increase the order only in a specific frequency band. It has been introduced.

FIG. 7B shows a state in which the concept of “sight” is introduced into the same visible graph as FIG. The visual acuity is a numerical value representing a range of nodes connected by an edge, and is represented by “N _vis ”. In the example shown in FIG. 7B, N _vis = 3. In this case, three nodes (right side) from the connection source node are used as connection destination nodes, and only the nodes that have a line of sight are connected by edges. Therefore, the degree of each node is 3 at maximum. In the example shown in FIG. 7A, N _vis = inf (Infinity: Infinity) is set because no limitation is imposed by visual acuity.

FIG. 8B shows a state in which the concept of “sight” is introduced to the same visible graph as FIG. The visual acuity is N _vis = 3.
As is clear from FIG. 7 and FIG. 8, by introducing the concept of “sight”, when a suddenly high value occurs or a state with a frequency lower than a specific frequency band occurs, the average It is possible to prevent the order from increasing extremely.

A specific visual acuity is set by the following equation (3).
Here, n is a natural number, _{f s} is the sampling frequency (Hz), f is to be detected the frequency (Hz).

The sampling frequency f _s is the number of data of the pressure fluctuation signal acquired per second. The frequency to be detected is the frequency at which the average order is to be set highest. Therefore, in Equation (3), the value of f _s / f is the number of data of the pressure fluctuation signal acquired in one cycle of the frequency to be detected, as shown in FIG. 9A. A value obtained by approximating the value of f _s / f by a power of 2 is defined as visual acuity N _vis . When the visual acuity N _vis is set using Equation (3), the average order does not decrease when the pressure fluctuation signal fluctuates at a frequency higher than the frequency to be detected (see FIG. 9B), but from the frequency to be detected. If the pressure fluctuation signal fluctuates at a lower frequency, the average order decreases (the increase in the average order can be suppressed) (see FIG. 9C). Further, in the formula (3), the visual acuity N _vis is approximated by a power of 2 because the peak frequency confirmed by the combustion vibration is not always constant. In addition, you may employ | adopt other forms other than the power of 2 for the approximation of visual acuity _Nvis .

  The unstable combustion state in the combustion chamber 11 includes combustion vibration and blow-off. FIG. 10 is a graph illustrating the relationship between the average order of complex networks and the equivalence ratio. In FIG. 10, the area where the equivalence ratio is smaller than about 0.51 is an area where blowout is likely to occur (hereinafter, also referred to as “blowout occurrence area”), and the area where the equivalent ratio is about 0.61 or more is This is a region where combustion vibration is likely to occur (hereinafter also referred to as “combustion vibration generation region”). Between the blow-off generation region and the combustion vibration generation region is a region where the combustion state is stable, and it is desirable to control the fuel supply amount so as to be maintained in this region. Therefore, the region is a region to be controlled (hereinafter also referred to as “control region”).

The pressure in the combustion chamber 11 fluctuates stably in a relatively low frequency band in the combustion vibration generation region, and fluctuates at a higher frequency in the control region. Furthermore, more complicated fluctuations occur at higher frequencies in the blow-off region. Therefore, in the present embodiment, when introducing the visual acuity N _vis as described above, the frequency in the region is set as the frequency f to be detected so that the average order in the combustion vibration generation region is high. By setting in this way, the average order in the combustion vibration generation region can be increased, and the average order in a region other than the combustion vibration generation region can be decreased.

  In the example shown in FIG. 10, the average order is high in the combustion region generation region where the equivalence ratio is about 0.61 or more, and the average order is low in the control region where the equivalence ratio is less than about 0.61, and both are clearly It can be seen that they are distinguished. In the control region, it can be seen that the average order decreases with a large change as the equivalence ratio decreases.

Returning to FIG. 2, the threshold setting unit 315 is for setting an average order threshold <k> _thre for determining whether blow-off occurs. In the present embodiment, the threshold value <k> _thre is stored in a hard disk of a computer or the like, and the threshold setting unit 315 reads the stored threshold value <k> _thre and sets it by storing it in the RAM as a set value. . The threshold setting unit 315 may receive information indicating the threshold <k> _thr from the user and store the threshold <k> _thr in the storage unit as a set value, or may be given from other software. The threshold information may be received and stored as a set value in the storage unit.

The blow-out occurrence estimation unit 316 is for estimating the occurrence of blow-out based on the average order <k> calculated by the average order calculation unit 314. As shown in FIG. 10, the average order <k> decreases as the combustion state in the combustion chamber 11 approaches the blowout occurrence region. Therefore, the blow-out occurrence estimation unit 316 compares the average order <k> calculated by the average order calculation unit 314 with the threshold <k> _thr set by the threshold setting unit 315, and calculates the average order <k>. _Is smaller than the threshold value <k> _thre , it is estimated that the occurrence of blow-off is approached.

  The secondary fuel flow rate calculation unit 317 is for calculating the secondary fuel flow rate based on the estimation result by the blowout occurrence estimation unit 316. As will be described later, the fuel supply path extending from the fuel tank 4 is branched in the middle, one being the main fuel supply path and the other being the secondary fuel supply path (see FIG. 1). The secondary fuel flow rate calculation unit 317 calculates the flow rate of the secondary fuel supplied from the secondary fuel supply path.

In the control region shown in FIG. 10, it is estimated that the occurrence of blow-out is closer as the average order <k> is smaller. Blow-out is a phenomenon that occurs when the equivalence ratio of the premixed gas is small. In order to eliminate the occurrence of blow-out, it is necessary to increase the equivalence ratio. Therefore, the secondary fuel flow rate calculation unit 317 increases the secondary fuel flow rate as the difference between the average order <k> and the threshold value <k> _thr increases. Specifically, the change amount of the secondary fuel flow rate is calculated according to the following equation (4), and the value changed by the change amount with respect to the secondary fuel flow rate at that time is calculated as the secondary fuel flow rate. Determine as target value.

ΔQ _{CH4, secondary} is the amount of change in the secondary fuel flow rate for 50 ms, α is a proportional constant, <k> is the average order obtained from the data of the pressure fluctuation signal for 50 ms, and <k> _thr is the average order Is the threshold value. The proportionality constant α can be set to α = 0.01, for example.

  The secondary fuel flow rate control unit 318 is for controlling the secondary fuel flow rate so as to approach the target value of the secondary fuel flow rate determined by the secondary fuel flow rate calculation unit 317. The secondary fuel flow rate control unit 318 outputs information indicating the target value to the mass flow controller 33 (see FIG. 1) provided in the middle of the secondary fuel supply path.

  The mass flow controller 32 is provided in the middle of the main fuel supply path. The mass flow controller 32 is connected to the signal analysis device 31 and controls the main fuel flow rate in the main fuel supply path in accordance with a control signal output from the signal analysis device 31.

  The mass flow controller 33 is provided in the middle of the secondary fuel supply path. The mass flow controller 33 is connected to the signal analysis device 31 and controls the secondary fuel flow rate to approach the target value according to a control signal (information indicating the target value) output from the signal analysis device 31.

  The fuel tank 4 stores methane as fuel. A fuel supply path extends from the fuel tank 4 and is connected to an intake portion 13 of the gas turbine model combustor 1.

  The compressor 5 supplies air containing oxygen which is an oxidant. An air supply path extends from the compressor 5 and is connected to the intake portion 13 of the gas turbine model combustor 1.

  In the middle of the air supply path, a mass flow controller 51 is provided (see FIG. 1). The mass flow controller 51 controls the flow rate of air supplied to the gas turbine model combustor 1 according to a given set value.

[2. Operation of observation equipment]
Hereinafter, the operation of the observation apparatus 100 will be described.
In response to an instruction from the signal analyzer 31, each of the mass flow controllers 32 and 33 controls the main fuel flow rate and the secondary fuel flow rate, whereby methane is supplied from the fuel tank 4. Air is supplied from the compressor 5, and the supply amount is controlled by the mass flow controller 51.

  Methane supplied from the fuel tank 4 and air supplied from the compressor 5 are taken into the intake section 13. The intake unit 13 mixes fuel and air to generate an air-fuel mixture. The generated air-fuel mixture is swirled by a swirler and introduced into the combustion chamber 11 to burn the air-fuel mixture.

  A pressure fluctuation in the combustion chamber is detected by the pressure transducer 2. The pressure fluctuation signal output from the pressure transducer 2 is amplified by the amplifier 21 and supplied to the signal analysis device 31.

  The signal analysis device 31 executes a combustion control process and outputs a control signal indicating the target value of the secondary fuel flow rate obtained thereby to the mass flow controller 33. The mass flow controller 33 controls the secondary fuel flow rate so as to approach the target value. In this way, the combustion state of the gas turbine model combustor 1 is feedback controlled.

  Hereinafter, the combustion control process by the signal analyzer 31 will be described. FIG. 11 is a flowchart showing the procedure of the combustion control process. First, a pressure fluctuation signal is acquired (step S1). In this process, the pressure fluctuation signal input unit 311 stores the input pressure fluctuation signal in the storage unit 312. Here, the time series data of the pressure fluctuation signal for 50 msec is stored in the storage unit 312. For example, when the sampling frequency of the pressure fluctuation signal is 5 kHz, 250 time-series data are stored in the storage unit 312.

Next, a complex network generation process is executed (step S2). In the complex network generation process, a visible graph is generated from the time-series data of the pressure fluctuation signal read from the storage unit 312, and a complex network composed of nodes and edges is generated by the above-described equation (2). Further, when connecting nodes by edges, the visual acuity N _vis obtained by the above-described equation (3) is taken into consideration.

  Next, an average order calculation process in the complex network is executed (step S3). In this calculation process, the number of nodes of the complex network generated in the previous step S2 and the number of edges (order) connected to each node are acquired, and the average order is calculated by the above-described equation (1).

Next, a secondary fuel flow rate calculation process is executed (step S4). FIG. 12 is a flowchart showing the procedure of the secondary fuel flow rate calculation process. First, in the secondary fuel flow rate calculation process, a threshold value <k> _thr of an average order <k> is set (step S41). In this process, the threshold setting unit 315 reads the threshold _{<k> thre} a hard disk or the like, sets the threshold value _{<k> thre} by storing in RAM. The threshold value <k> can be set to <k> _thr = 8.87, for example, in the example shown in FIG.

Next, the secondary fuel flow rate _{QCH4, secondary} at that time is calculated (step S42). In this processing, the secondary fuel flow rate calculation unit 317 calculates the secondary fuel flow rate _{QCH4, secondary} at that time using the set value of the mass flow controller 33 or the like.

Next, a change amount ΔQ _{CH4, secondary} of _{the secondary} fuel flow rate is calculated (step S43). In this process, the blowout occurrence estimation unit 316 estimates the occurrence of _blowout by comparing the average order <k> and the threshold value <k> _thre, and calculates the average order <k> and the threshold value <k> _thre . The change amount ΔQ _{CH4, secondary} of the _secondary fuel flow rate according to the difference is calculated according to the above-described equation (4).

Next, the change amount ΔQ _{CH4, secondary} is reflected in the current secondary fuel flow rate Q _{CH4, secondary} . That is, as the target value of the secondary fuel flow rate _{, the secondary} fuel flow rate Q _{CH4, secondary} and the change amount ΔQ _{CH4, secondary} of the secondary fuel flow rate are added to obtain the target value of the _secondary fuel flow rate Q _{CH4, secondary} . Is calculated (step S44). When this process ends, the process is returned to the calling address of the secondary fuel flow rate calculation process in the main routine.

  Returning to FIG. 11, next, the secondary fuel flow rate control process is executed (step S5). In this process, the secondary fuel flow rate control unit 318 transmits a control signal indicating the target value calculated in step S4 to the mass flow controller 33, thereby controlling the secondary fuel flow rate so as to approach the target value.

  When the process of step S5 ends, it is determined whether or not to end the combustion control process (step S6). In this process, it is determined that the combustion control process is to be ended when a predetermined end condition is satisfied, for example, when the signal analysis device 31 is instructed to stop the operation. If it is determined not to end the combustion control process (NO in step S6), the process is returned to step S1, whereby the processes of steps S1 to S6 are repeatedly executed every 50 msec. On the other hand, when it is determined in step S6 that the combustion control process is to be ended (YES in step S6), the combustion control process is ended.

  As described above, the combustion control process is repeatedly executed every 50 msec. That is, as shown in FIG. 3, a complex network is generated from the pressure fluctuation signal for a period of 50 msec, and the average order <k> is calculated. Therefore, the combustion state of the gas turbine model combustor 1 is controlled every 50 msec.

[3. Evaluation test results]
The inventors of this application performed the performance evaluation test of the observation apparatus 100 which concerns on this embodiment. 13 and 14 are graphs showing the results of this performance evaluation test.
In the performance evaluation test shown in FIG. 13, as shown in the upper graph, the flow rate of the main fuel is decreased at a constant rate from the initial value of about 8 L / min to about 6 L / min, and then the flow rate of about 6 L / min. Maintained. Since the limit of occurrence of blow-off is when the total flow rate of fuel is about 6.5 L / min, the main fuel flow rate is changed until it falls below this limit.

  In the observation apparatus 100 according to the present embodiment, feedback control is applied when the main fuel flow rate reaches about 6.7 L / min, and the supply of secondary fuel is started as shown in the middle graph. As a result, as shown in the lower graph, the equivalence ratio was maintained at a value exceeding about 0.48, which is the blowout occurrence limit.

  In the performance evaluation test shown in FIG. 14, as shown in the upper graph, the flow rate of the main fuel is decreased at a constant rate from the initial value of about 8 L / min to about 6 L / min, and then constant to about 8 L / min. Increased at a ratio of Also in this case, the limit of occurrence of blow-off is when the total flow rate of the fuel is about 6.5 L / min. Therefore, the main fuel flow rate is changed until it falls below this limit.

  In the observation apparatus 100 according to the present embodiment, feedback control is applied when the main fuel flow rate reaches approximately 6.9 L / min, and the supply of secondary fuel is started as shown in the middle graph. And while the main fuel flow rate was decreasing, the secondary fuel flow rate continued to increase, and when the main fuel flow rate turned from increase to decrease, the secondary fuel flow rate changed from increase to decrease. As a result, as shown in the lower graph, the equivalence ratio was maintained at a value exceeding about 0.48, which is the blowout occurrence limit. When the main fuel flow rate recovered to the original value (about 8 L / min), the supply of secondary fuel was stopped.

In this type of combustor 1, in order to reduce the generation of carbon dioxide and NO _X, it is desirable to as much as possible dilute the premixed gas. However, if the premixed gas is too dilute, blow-off occurs, and a stable combustion state cannot be maintained. From the above evaluation test results, it can be seen that the observation apparatus 100 according to the present embodiment maintains a low equivalent ratio in the vicinity of the blowout occurrence limit while avoiding the occurrence of blowout. Thus, according to the observation apparatus 100 according to the present embodiment, stable lean premixed combustion can be maintained.

In the present embodiment, as shown in FIG. 10, the average order <k> greatly changes between about 6 and about 14 with a change in the equivalence ratio in the control region, and changes even when the blow-off occurrence region is approached. The amount remains large. Therefore, by setting the value of the average order <k> in the vicinity of the blow-out occurrence region as the threshold value <k> _thr , the occurrence of blow-out can be stably estimated.

  Further, as in the example shown in FIG. 10, it can be seen that the value of the average order <k> changes significantly in the vicinity of the boundary between the combustion vibration generation region and the control region. By using such an average order <k>, it is also possible to estimate whether combustion vibration has occurred.

  The observation apparatus 100 according to the present embodiment generates a complex network using a visible graph based on time-series data of pressure fluctuation signals, and generates combustion instability according to an average order <k> that is a characteristic amount of the complex network. Estimated. Therefore, the operation can be performed by a very simple algorithm. Therefore, the calculation speed can be increased, and the combustion state of the combustor can be observed with high immediacy and good responsiveness. Further, the signal analyzer 31 is not required to have a high capability.

  Moreover, the observation apparatus 100 of this embodiment can reduce the setting of the parameter used for a calculation. Specifically, in the technique described in Patent Document 2, the number of parameters set is 7, but in the present embodiment, only two parameters, the number of time-series data and visual acuity, are set. Therefore, parameter setting is easy.

  In addition, we introduced the concept of “sight” when creating complex networks, and appropriately limited the range of nodes connected by edges, so the effects of sudden fluctuations in time series data and low frequency fluctuations Can be reduced, the amount of calculation can be reduced efficiently, and the calculation speed can be further increased.

  Moreover, according to the observation apparatus 100 according to the present embodiment, when it is estimated that blow-off occurs, the secondary fuel flow rate is controlled to avoid occurrence of blow-out. Thereby, generation | occurrence | production of blowing-out can be suppressed and the stable combustion state can be maintained in the gas turbine model combustor 1. FIG.

  Moreover, according to the observation apparatus 100 according to the present embodiment, the pressure fluctuation signal for 50 msec is constantly stored by updating the pressure fluctuation signal stored in the storage unit 312 with the newly input pressure fluctuation signal. Thus, the average order is calculated every 50 msec using the pressure fluctuation signal for 50 msec stored in the storage unit 312. For this reason, the combustion state can be continuously observed with high immediacy, and the secondary fuel flow rate can be appropriately controlled.

[4. Other Embodiments]
In the above-described embodiment, the configuration for observing the combustion state in the gas turbine model combustor 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 the target. In addition, for the time series measurement signals of the distortion of the rotor blades (fans and compressors) of the gas turbine engine and the pressure fluctuations before and after the rotor blades, the feature amount of the complex network is calculated, and flutter, turning stall, It can also be set as the structure which estimates generation | occurrence | production of a surge. In addition, for industrial plants, complex network features are also available for unstable pressure fluctuation signals of gas-liquid two-phase flow in an evaporation pipe and fluid temperature fluctuation signals in a pipe junction where two fluids with different temperature differences are mixed. To calculate the occurrence of other physical phenomena such as damage to the boiler's evaporation pipe and health monitoring of damage to the piping due to thermal striping.

  In the above-described embodiment, the configuration in which the occurrence of blow-out is estimated based on the feature amount of the complex network and the combustion state is controlled to avoid the occurrence of blow-out has been described. However, the present invention is not limited to this. It is not something. For example, the occurrence of blow-off is estimated based on the feature amount of the complex network, but a configuration in which the combustion state is not controlled may be employed. The same applies to the configuration for estimating the occurrence of other physical phenomena.

  In the above-described embodiment, the configuration for estimating the occurrence of blow-out as the combustion instability in the gas turbine model combustor has been described. However, the present invention is not limited to this. A configuration may be adopted in which the occurrence of combustion vibration is estimated based on the characteristic amount of the complex network. In this case, in order to avoid the occurrence of combustion vibration, the main fuel flow rate or the secondary fuel flow rate may be controlled based on the characteristic amount of the complex network.

  In the above-described embodiment, the configuration is described in which the combustion state is controlled so as to avoid the occurrence of blow-out by controlling the secondary fuel flow rate, but is not limited thereto. By controlling the main fuel flow rate based on the characteristic amount of the complex network, the occurrence of blow-out may be avoided, or by controlling the air flow rate, occurrence of blow-out may be avoided. Good. In addition, the main fuel supply path and the secondary fuel supply path are not provided separately, but only one fuel supply path is provided, and the fuel flow rate in the fuel supply path is controlled based on the characteristic amount of the complex network. Thus, it is possible to adopt a configuration that avoids occurrence of blow-out.

  In the above-described embodiment, the configuration is described in which the pressure fluctuation signal for 50 msec is stored in the storage unit, and the feature amount is repeatedly calculated every 50 msec while updating the pressure fluctuation signal in the storage unit. It is not limited to this. For example, a pressure fluctuation signal during a period other than 50 msec, such as 100 msec, may be stored in the storage unit, and a feature amount may be calculated using the signal.

  Alternatively, a pressure fluctuation signal for a period of 50 msec or longer (for example, 1 sec) may be stored in the storage unit, and a pressure fluctuation signal for the latest predetermined period (for example, 50 msec) may be used for calculating the feature amount. Good.

  Further, instead of repeatedly calculating the feature amount every 50 msec, for example, the feature amount may be repeatedly calculated every period other than 50 msec, such as 10 msec. However, in order to appropriately control the combustion state with high immediacy, the configuration is such that the combustion state is controlled by calculating the feature amount at a cycle of 10 μsec to 300 msec according to the characteristic time of the phenomenon to be controlled. It is preferable.

  The feature amount of the complex network is not limited to the average order, and the average inter-vertex distance, the cluster coefficient, the mode value of the order distribution, the frequency of the mode value of the order distribution, or the like may be used. However, since the average order can be calculated simply by acquiring the number of edges connected to each node, there is an advantage that the calculation load is smaller than that of other feature amounts and the calculation can be performed at higher speed.

  The average inter-vertex distance is the average value of the minimum number of edges from node to node. The amount of variation in the average vertex distance can be reduced by applying the visual acuity of the above embodiment. The cluster coefficient is obtained by dividing the number of edges between three adjacent nodes by the maximum number of edges that can be taken between the three nodes. The cluster coefficient has a property similar to the average order in that it increases when nodes are closely connected. The mode value of the degree distribution is the order that is the mode value among the orders that are the number of edges connected to the node. The frequency of the mode value of the degree distribution is the frequency at which the order of the mode value appears. When the time-series data fluctuates with a sine wave, the mode value of the order distribution depends on the number of data for one wavelength of the sine wave, and the frequency becomes remarkably high, resulting in a biased degree distribution. . Therefore, the order distribution may change significantly more than the average order. The mode value and frequency of the degree distribution can be applied to time series data obtained from the Lorentz equation.

  In the above-described embodiment, the signal analysis device 31 is realized by a computer. However, the present invention is not limited to this. For example, hardware capable of executing combustion control processing may be configured by an ASIC, FPGA, or the like, and this may be used as a signal analysis device. The signal analysis device 31 may not be configured by one computer, but may be a distributed system including a plurality of computers by performing combustion control processing by a plurality of computers by distributed processing.

1: Gas turbine model combustor 2: Pressure transducer (detector)
3: Control unit 31: Signal analysis device 100: Observation device 313: Complex network generation unit 314: Average degree calculation unit 316: Blowout occurrence estimation unit _Nvis : Visual acuity

Claims (9)

A detection unit for detecting a physical quantity of a physical phenomenon that changes over time;
A generation unit that sets a physical quantity detected at each time as a node, and generates a complex network in which a plurality of nodes are connected by an edge according to the following conditions (A) and (B):
A calculation unit for obtaining a predetermined feature amount in the complex network;
An observation device comprising: an estimation unit that estimates a state of a physical phenomenon based on the feature amount.
(A): a node between which is set by the physical quantity detection time is adjacent (B): and the time t _a detected physical quantity y _a, a physical quantity y _b detected at time t _b after time t _a , the node between the physical quantity y _c detected at time t _c between time t _a and the time t _b is in the case which satisfies the following equation, is set by the physical quantity y _a physical quantity y _b
  The observation apparatus according to claim 1, wherein the generation unit limits a range of nodes connected by edges in the complex network based on a predetermined rule.   The observation apparatus according to claim 2, wherein the range of the node is set based on the number of physical quantity detections per change period of the physical quantity when a predetermined physical phenomenon occurs.   The observation apparatus according to claim 1, wherein the feature amount is an average order of the complex network.   The observation apparatus according to claim 4, wherein the estimation unit estimates a state of a physical phenomenon based on a comparison between the average order and a predetermined threshold value.   The observation apparatus according to claim 1, wherein the physical phenomenon is combustion of fuel in a combustor, and the physical quantity is pressure in the combustor.   The observation apparatus according to claim 6, wherein the estimation unit estimates occurrence of instability of combustion in the combustor.   The observation device according to claim 7, wherein the estimation unit estimates occurrence of blow-off as instability of combustion in the combustor. Detecting a physical quantity of a physical phenomenon that changes over time;
A step of setting a detected physical quantity at each time as a node and generating a complex network in which a plurality of nodes are connected by an edge according to the following conditions (A) and (B):
Obtaining a predetermined feature amount in the complex network;
Estimating the state of a physical phenomenon based on the feature amount.
(A): a node between which is set by the physical quantity detection time is adjacent (B): and the time t _a detected physical quantity y _a, a physical quantity y _b detected at time t _b after time t _a , the node between the physical quantity y _c detected at time t _c between time t _a and the time t _b is in the case which satisfies the following equation, is set by the physical quantity y _a physical quantity y _b