Observation apparatus, observation method, and non-transitory computer readable medium storing a program

Abstract

[Object] To observe the sign or occurrence of an unstable operation of a turbo-machine. [Solving Means] An observation apparatus 1 includes: a detection unit 10 including one or two or more sensors 11, 12 that are disposed in a turbo-machine 2, are highly time responsive, and observe unsteady fluctuations of the turbo-machine 2; a computation unit 20 that output signals from the one or two or more sensors 11, 12 every moment, stores time series data for a predetermined period, and calculates in real time a parameter for detecting an unstable operation of the turbo-machine; and a determination unit 30 that compares the parameter for detecting the unstable operation with a predetermined threshold and outputs in real time a determination result of a sign or occurrence of the unstable operation.

Claims

1. An observation apparatus, comprising: a detection unit including two or more sensors that are disposed in a turbo-machine, are highly time responsive, and observe unsteady fluctuations of the turbo-machine; a computation unit that output signals from the two or more sensors every moment, stores time series data for a predetermined period, and calculates in real time a parameter for detecting an unstable operation of the turbo-machine; a determination unit that determines a sign or occurrence of the unstable operation on a basis of the parameter for detecting the unstable operation; and a control unit that outputs, when the determination unit outputs a determination result of the sign or occurrence of the unstable operation, a signal for changing an operation condition for an operation control apparatus of the turbo-machine and/or a signal for warning of an operation of the turbo-machine, wherein the computation unit calculates the parameter for detecting the unstable operation by quantitatively evaluating randomness and a recurrence change on a basis of the time series data, wherein the detection unit includes two or more types of detection units, the computation unit includes two or more types of computation units, and the determination unit includes two or more types of determination units, wherein the two or more types of computation units calculate two or more types of parameters for detecting the unstable operation by quantitatively evaluating randomness and the recurrence change on the basis of the time series data, and wherein the two or more types of determination units determine the sign or occurrence of two or more types of the unstable operation at a same time.

2. The observation apparatus according to claim 1, wherein the two or more sensors are disposed on at least one of a rotating unit, a stationary unit, an inside of a flow channel, or a wall surface in contact with the flow channel in the turbo-machine.

3. The observation apparatus according to claim 1, wherein the computation unit calculates the parameter as a sample entropy that is an index for quantitatively evaluating the randomness of the time series data.

4. The observation apparatus according to claim 3, wherein provided that the time series data is expressed as {x(t.sub.i)}, i=1, 2, . . . , N, the time series data {x(t.sub.i)} is embedded in phase spaces of D and D+1 dimensions, and conditional probability that a point that was nearby in the D dimension is also nearby in the D+1 dimension is defined as a negative natural logarithm, and provided that the sample entropy is denoted by SE, the sample entropy SE is calculated by the following equation $S_E=-\log \frac{\underset{i=1,i\ne j}{\overset{N-D}{.Math.}}\Theta \left(r-d\left[X_{D+1}\left(i\right),X_{D+1}\left(j\right)\right]\right)}{\underset{i=1,i\ne j}{\overset{N-D+1}{.Math.}}\Theta \left(r-d\left[X_D\left(i\right),X_D\left(j\right)\right]\right)}$ where r denotes a predetermined threshold
d[X.sub.D(t.sub.i),X.sub.D(t.sub.j)]=max|x(t.sub.i+k)−x(t.sub.j+k)|
X.sub.D(t.sub.i)=(x(t.sub.i),x(t.sub.i+1),x(t.sub.i+2), . . . ,x(t.sub.i+D−1)).

5. The observation apparatus according to claim 1, wherein the computation unit calculates the parameter as a sample entropy considering a multi-scale property that is an index for performing coarse graining on the time series data and quantitatively evaluating the randomness of the time series data after the coarse graining.

6. The observation apparatus according to claim 5, wherein provided that the time series data is denoted by x(t.sub.i), a time average of the time series data x(t.sub.i) is determined by non-overlapping average using the following equation and time series data y(t.sub.j) is obtained $y\left(t_j\right)=\frac{1}{s_f}\stackrel{{\mathrm{js}}_f}{\underset{i=\left(j-1\right)s_f+1}{.Math.}}x\left(t_i\right)$ where x(t.sub.i): Time series s.sub.f: Scaling factor y(t.sub.j): Coarse-grained time series, and the sample entropy is calculated by using the time series data y(t.sub.j).

7. The observation apparatus according to claim 1, wherein the computation unit calculates the parameter as recurrence plots that are an index for embedding the time series data in a phase space and visualizing a correlation between respective points of the time series data in the phase space.

8. The observation apparatus according to claim 7, wherein provided that the time series data is denoted by x(t.sub.i), computation according to the following equation is performed with respect to the correlation between the respective points of the time series data in the phase space and a result of computation is plotted for obtaining the recurrence plots
X(t.sub.i)=(x(t.sub.i),x(t.sub.i+τ), . . . ,x(t.sub.i+(D−1)τ))
R.sub.ij=Θ(ε−∥x(t.sub.i)−x(t.sub.j)∥) i,j=1,2,3, . . . ,N.sub.P where θ: Heaviside function ε; threshold of distance between position vectors N.sub.P: total number of data points in phase space D: dimension of phase space τ: delay time.

9. The observation apparatus according to claim 8, wherein provided that an index for determining the sign or occurrence of the unstable operation is denoted by DET, the determination unit calculates the index DET in the obtained recurrence plots in accordance with the following equation $\mathrm{DET}=\frac{\underset{l=l_{\min }}{\overset{N_p}{.Math.}}\mathrm{lP}\left(l\right)}{\underset{l=1}{\overset{N_p}{.Math.}}\mathrm{lP}\left(l\right)}$ where l: length of diagonal line in recurrence plots l.sub.min: minimum length defined as diagonal line in recurrence plots P(l): frequency distribution function of diagonal line having length l in recurrence plots.

10. The observation apparatus according to claim 1, wherein the computation unit calculates the parameter as a permutation entropy that is an index for quantitatively evaluating the randomness of the time series data.

11. The observation apparatus according to claim 10, wherein the time series data is classified into predetermined permutation patterns and the permutation entropy is calculated by applying existence probability of each of the permutation patterns to Shannon's information entropy.

12. The observation apparatus according to claim 11, wherein the Shannon's information entropy is expressed by the following equation using a discrete probability distribution p of a random variable of an event $s=-\underset{i=1}{\overset{N}{.Math.}}p{\log }_{2}p$ where N=D! D: dimension of phase space, and provided that the permutation entropy is denoted by h.sub.p, the permutation entropy h.sub.p is calculated in accordance with the following equation $0\le h_p=\frac{-\stackrel{D!}{\underset{i=1}{.Math.}}p\left(i\right){\log }_{2}p\left(i\right)}{{\log }_{2}D!}\le 1.$

13. The observation apparatus according to claim 1, wherein the determination unit compares the parameter for detecting the unstable operation with a predetermined threshold and outputs in real time the determination result of the sign or occurrence of the unstable operation.

14. An observation method, comprising: disposing in a turbo-machine two or more detection units, each including two or more sensors that are highly time responsive to the turbo-machine and observe unsteady fluctuations of the turbo-machine; inputting output signals from the two or more sensors of each detection unit every moment, storing time series data for a predetermined period, and calculating, by two or more computation units, in real time two or more parameters for detecting an unstable operation of the turbo-machine by quantitatively evaluating randomness and a recurrence change on a basis of the time series data; and determining, by two or more detection units, a sign or occurrence of two or more types of the unstable operation at a same time on a basis of the two or more parameters; and when a determination unit of the two or more determination units outputs a determination result of the sign or occurrence of the unstable operation, outputting, by a control unit, a signal for changing an operation condition for an operation control apparatus of the turbo-machine and/or a signal for warning of an operation of the turbo-machine.

15. A non-transitory computer readable medium storing a program that causes a computer to execute: a step of inputting output signals from two or more sensors, of each detection unit of two or more detection units disposed in a turbo-machine, storing time series data for a predetermined period, and calculating, by two or more computation units, in real time two or more parameters for detecting an unstable operation of the turbo-machine by quantitatively evaluating randomness and a recurrence change on a basis of the time series data, each sensor being highly time responsive to the turbo-machine and being configured to observe unsteady fluctuations of the turbo-machine; a step of determining, by two or more detection units, a sign or occurrence of two or more types of the unstable operation at a same time on a basis of the two or more parameters; and when a determination unit of the two or more determination units outputs a determination result of the sign or occurrence of the unstable operation, outputting, by a control unit, a signal for changing an operation condition for an operation control apparatus of the turbo-machine and/or a signal for warning of an operation of the turbo-machine.

Description

BRIEF DESCRIPTION OF DRAWINGS

(1) FIG. 1 A block diagram showing an observation apparatus according to an embodiment of the present invention.

(2) FIG. 2 A schematic diagram showing an example in which sensors according to the embodiment of the present invention are disposed in a turbo-machine.

(3) FIG. 3 A graph showing an arrangement example of time series data for describing a sample entropy.

(4) FIG. 4 A graph showing an example of the arrangement of the time series data in a case where D=1 in the sample entropy.

(5) FIG. 5 A graph showing an example of the arrangement of the time series data in a case where D=2 in the sample entropy.

(6) FIG. 6 An explanatory diagram of a method of obtaining time series data obtained by coarse graining considering a multi-scale property.

(7) FIG. 7 A graph showing a result of computation using strain fluctuations of blades of the turbo-machine in a sample entropy considering the multi-scale property.

(8) FIG. 8 A graph showing an example of time series data of pressure fluctuations for describing recurrence plots.

(9) FIG. 9 A graph showing an example in which the time series data shown in FIG. 8 is embedded in a phase space.

(10) FIG. 10 A graph showing respective points in the phase space.

(11) FIG. 11 A graph representing a correlation between the respective points in the phase space shown in FIG. 10 as recurrence plots.

(12) FIG. 12 A graph showing a relationship between each flow rate and DET of the recurrence plots in this embodiment.

(13) FIG. 13 A relationship diagram of recurrence plots at a flow rate q=6 kg/s.

(14) FIG. 14 A relationship diagram of recurrence plots at a flow rate q=8.5 kg/s.

(15) FIG. 15 A relationship diagram of recurrence plots at a flow rate q=9.5 kg/s.

(16) FIG. 16 An explanatory diagram for capturing flutter on the basis of a relationship between each flow rate and the DET of the recurrence plots in this embodiment.

(17) FIG. 17 A graph showing an example of the time series data for describing a permutation entropy.

(18) FIG. 18 A diagram classifying the time series data shown in FIG. 17 into permutation patterns.

(19) FIG. 19 A graph showing existence probability of the permutation patterns shown in FIG. 18.

(20) FIG. 20 An explanatory diagram for capturing flutter on the basis of the permutation entropy.

(21) FIG. 21 A graph showing an example of a change over time of the permutation entropy.

(22) FIG. 22 A block diagram showing an observation apparatus according to another embodiment of the present invention.

(23) FIG. 23 A block diagram showing an observation apparatus according to still another embodiment of the present invention.

MODE(S) FOR CARRYING OUT THE INVENTION

(24) Hereinafter, an embodiment of the present invention will be described with reference to the drawings.

(25) <Configuration of Observation Apparatus>

(26) FIG. 1 is a block diagram showing an observation apparatus according to the embodiment of the present invention.

(27) As shown in FIG. 1, an observation apparatus 1 includes a detection unit 10, a computation unit 20, a determination unit 30, and a control unit 40.

(28) The detection unit 10 includes one or two or more sensors that are disposed in a turbo-machine 2, are highly time responsive, and observe unsteady fluctuations of the turbo-machine 2.

(29) As shown in FIG. 2, the detection unit 10 includes a sensor 11 attached to a blade 2a which is the rotating unit of the turbo-machine 2, a sensor 12 attached to a wall surface 2b which is the stationary unit facing a leading end of the blade 2a, and the like. Such a sensor that is the detection unit 10 may be disposed in a flow channel or on a wall surface in contact with the flow channel.

(30) The sensor 11 is constituted of, for example, a strain gauge that detects strain of the blade in real time and the sensor 12 is constituted of, for example, an unsteady pressure sensor that detects the pressure of the fluid in real time. In this embodiment, these sensors 11 and 12 are for observing flutter, which is one of the unstable operations. The sensor 12 may be disposed in a flow channel or on a wall surface in contact with the flow channel, for example. In order to observe stall and surge, which are unstable operations, it is sufficient to dispose sensors in a similar manner.

(31) The calculation unit 20 inputs output signals from the sensor 11 and the sensor 12 every moment, stores time series data for a predetermined period, and quantitatively evaluates randomness and a recurrence change on the basis of the time series data, to thereby calculate a parameter for detecting an unstable operation of the turbo-machine 2 in real time. For example, the computation unit 20 inputs an output signal from the sensor 11 every moment, stores time series data for a predetermined period, and calculates a parameter for detecting flutter in real time.

(32) The determination unit 30 compares the parameter for detecting the unstable operation with a predetermined threshold and outputs a determination result of the sign or occurrence of the unstable operation in real time.

(33) When the determination unit 30 outputs the determination result of the sign or occurrence of the unstable operation, the control unit 40 outputs a signal for changing the operation condition to an operation control apparatus 3 of the turbo-machine 2. Moreover, when the determination unit 30 outputs the determination result of the sign or occurrence of the unstable operation, the control unit 40 outputs a signal for warning a reporting unit 4 of an operation of the turbo-machine 2.

(34) When the operation control apparatus 3 receives the signal for changing the operation condition, the operation control apparatus 3 controls the turbo-machine 2 to stop the operation of the turbo-machine 2, for example.

(35) When the reporting unit 4 receives the signal associated with the warning, the reporting unit 4 supplies an alarm signal for a pilot or operator to perform manual control for an aircraft, for example.

(36) Here, as a method of calculating the detection parameter in the computation unit 20, it is effective to use a method of calculating the detection parameter by using an index of a sample entropy, a sample entropy considering a multi-scale property, recurrence plots, or a permutation entropy. Hereinafter, the method of calculating the detection parameter by using these indices will be described.

(37) (Sample Entropy)

(38) The sample entropy refers to an index for quantitatively evaluating the randomness of the time series data. Specifically, time series data {x(t.sub.i)}, i=1, 2, . . . , N is embedded in phase spaces of D and D+1 dimensions, and the conditional probability that a point that was nearby in the D dimension is also nearby in the D+1 dimension is defined as a negative natural logarithm.

(39) A sample entropy SE is as follows.

(40) $S_E=-\log \frac{\underset{i=1,i\ne j}{\overset{N-D}{.Math.}}\Theta \left(r-d\left[X_{D+1}\left(i\right),X_{D+1}\left(j\right)\right]\right)}{\underset{i=1,i\ne j}{\overset{N-D+1}{.Math.}}\Theta \left(r-d\left[X_D\left(i\right),X_D\left(j\right)\right]\right)}$

(41) Here, the following equations are established.
d[X.sub.D(t.sub.i),X.sub.D(t.sub.j)]=max|x(t.sub.i+k)−x(t.sub.j+k)|
X.sub.D(t.sub.i)=(x(t.sub.i),x(t.sub.i+1),x(t.sub.i+2), . . . ,x(t.sub.i+D−1))

(42) Then, for example, as shown in FIG. 3, θ(⋅)=1 in a case where X.sub.D(t.sub.j) exists in the D-dimensional cube centered on X.sub.D(t.sub.i) and θ(⋅)=0 in a case where X.sub.D(t.sub.j) does not exist in the D-dimensional cube centered on X.sub.D(t.sub.i).

(43) Here, for example, as shown in FIG. 4, provided that D=1 and the point ⋄ is used as a reference, two points Δ and two points ∇ are counted.

(44) As shown in FIG. 5, provided that D=2 and the point ⋄ is used as a reference, only two points Δ are counted. The points ∇ are not counted in a case where D=2 because the points ∇ are moved outside the D-dimensional cube by extending the dimensionality.

(45) A similar procedure is performed at all discrete points of the time series data and the sample entropy S.sub.E is calculated.

(46) In this embodiment, settings are performed such that a threshold r is 0.15 times as large as the standard deviation and D=2. By setting the threshold as appropriate, it is possible to capture the sign and occurrence of the unstable operation phenomenon of the turbo-machine 2.

(47) (Sample Entropy Considering Multi-Scale Property)

(48) The sample entropy considering the multi-scale property refers to an index for performing coarse graining on the time series data and using the sample entropy. Specifically, the time average of the time series data x(t.sub.i) is determined by non-overlapping average as follows and new time series data y(t.sub.j) as shown in FIG. 6 is obtained.

(49) $y\left(t_j\right)=\frac{1}{s_f}\stackrel{{\mathrm{js}}_f}{\underset{i=\left(j-1\right)s_f+1}{.Math.}}x\left(t_i\right)$
x(t.sub.i): Time series
s.sub.f: Scaling factor
y(t.sub.j): Coarse-grained time series

(50) Then, the sample entropy S.sub.E is calculated by substituting this new time series data into the defined equation above.

(51) With the sample entropy considering the multi-scale property, it is possible to know influences of different time scales by coarse graining.

(52) In this embodiment, computation was performed by using circumferential strain fluctuations 6 of the blade 2a detected from the sensor 11 as x. The example is shown in FIG. 7.

(53) As shown in A of FIG. 7, S.sub.E decreases in the low frequency region after S.sub.f=15 at the flow rate q=9.0 kg/s. S.sub.f (=15)×(second-order natural frequency) substantially coincides with the sampling frequency and captures the frequency characteristics of the second-order natural frequency. Therefore, it is possible to capture the sign of flutter by detecting it.

(54) Moreover, S.sub.E is low in the entire region at q=9.5 kg/s. Therefore, it is possible to capture the sign of flutter by detecting it.

(55) (Recurrence Plots)

(56) The recurrence plots refer to an index for visualizing the correlation between the respective points in the phase space. For example, first of all, the time series of pressure fluctuations shown in FIG. 8 is embedded in the phase space as shown in FIG. 9. Next, the correlation between the respective points in the phase space shown in FIG. 10 is plotted as shown in FIG. 11 by performing computation according to the following equation.

(57) Here,
X(t.sub.i)=(x(t.sub.i),x(t.sub.i+τ), . . . ,x(t.sub.i+(D−1)τ))
R.sub.ij=Θ(ε−∥x(t.sub.i)−x(t.sub.j)∥) i,j=1,2,3, . . . ,N.sub.P

(58) In the recurrence plots, an index DET representing determinism is calculated in accordance with the following equation.

(59) $DET=\frac{\underset{l=l_{\min }}{\overset{N_p}{.Math.}}\mathrm{lP}\left(l\right)}{\underset{l=1}{\overset{N_p}{.Math.}}\mathrm{lP}\left(l\right)}$

(60) In the equation above,

(61) θ: Heaviside function

(62) ε: threshold of distance between position vectors

(63) N.sub.P: total number of data points in phase space

(64) D: dimension of phase space (D=5 in this embodiment)

(65) τ: delay time (determined based on mutual information amount in this embodiment)

(66) l: length of diagonal line

(67) l.sub.min: minimum length defined as diagonal line

(68) P(l): frequency distribution function of diagonal line having length l.

(69) The recurrence plots relationship at each flow rate is shown in FIGS. 12 to 15.

(70) Comparing FIG. 13 with FIG. 14, it can be seen that the recurrence points are increased in the case where q=8.5 kg/s shown in FIG. 14 as compared with the case where q=6 kg/s shown in FIG. 13. Moreover, in the case where q=9.5 kg/s shown in FIG. 15, strong periodicity is observed with the occurrence of flutter. Thus, it can be understood that it is possible to capture the sign and occurrence of flutter by using the recurrence plots.

(71) In FIG. 16, results of Stc0, Stc1, and Stc2 in which flutter is developed are analyzed. Stc0, Stc1, and Stc2 are channel names of a strain gauge that detects the strain of the blade 2a in real time.

(72) As shown in C of FIG. 16, the above-mentioned index DET increases at q=8.5 kg/s before the occurrence of flutter. It is possible to detect the sign of flutter by detecting such a slight increase in DET. Moreover, as q increases thereafter, the periodicity of the waveform increases, such that the DET also increases (D of FIG. 16). By detecting it, it is possible to capture the occurrence of flutter.

(73) (Permutation Entropy)

(74) The permutation entropy refers to an index for quantitatively evaluating the randomness of the time series data. The time series data shown in FIG. 17 is classified into predetermined permutation patterns as shown in FIG. 18 and existence probability p of each of permutation patterns is obtained as shown in FIG. 19. The permutation entropy is calculated by applying p to the Shannon's information entropy. In general, the Shannon's information entropy is expressed by the following equation using a discrete probability distribution p of the random variable of the event. At this time, N=D!.

(75) $s=-\underset{i=1}{\overset{N}{.Math.}}p{\log }_{2}p$

(76) Where the permutation entropy is normalized by maximal entropy (=log.sub.2D!). In other words, a permutation entropy h.sub.p is determined by calculation as follows.

(77) $0\le h_p=\frac{-\stackrel{D!}{\underset{i=1}{.Math.}}p\left(i\right){\log }_{2}p\left(i\right)}{{\log }_{2}D!}\le 1$

(78) Here, a permutation entropy hp means more random as it is closer to 1 and means more periodic as it is closer to 0.

(79) Processed results at Stc0 to Stc8 are shown in FIG. 20. Stc0 to Stc8 are respectively the channel names of the strain gauges that detect the strains of the blade 2a in real time. From E and F in FIG. 20, the permutation entropy decreases because the periodicity of the waveform increases as q increases. It can be seen that the occurrence of flutter can be captured by detecting it.

(80) FIG. 21 shows a change over time in the permutation entropy.

(81) In FIG. 21,

(82) Q: change in air flow rate over time

(83) ε: change in strain fluctuations over time

(84) ε.sub.rms: change in root mean square of strain fluctuations over time.

(85) It can be seen from G of FIG. 21 that the permutation entropy h.sub.p decreases to ≈0.85 to 0.7 at t≈8.2 s before the ε.sub.rms increases rapidly. It can be seen from H of FIG. 21 that the permutation entropy h.sub.p decreases to ≈0.85 to 0.45 at t≈10 s when ε.sub.rms increases rapidly.

(86) Therefore, it is understood that although it is difficult to capture the sign and occurrence of flutter in ε.sub.rms, it is possible to capture the sign and occurrence of flutter by detecting a change in permutation entropy.

(87) <Others>

(88) The present invention is not limited to the above-mentioned embodiments and can be implemented as various modifications and applications without departing from the technical concept of the invention. The scope of such implementation is also encompassed in the technical scope of the present invention.

(89) For example, as shown in FIG. 22, an observation apparatus 100 may include two or more detection units 10, two or more computation units 20, and two or more determination units 30. By changing the system for observation of the unstable operation into a redundant system in this manner, the reliability can be enhanced.

(90) Moreover, as shown in FIG. 23, the observation apparatus 100 may include two or more types of detection units 10a to 10d, two or more types of computation units 20a to 20d, and two or more types of determination units 30a to 30d. Accordingly, it is possible to detect more kinds of unstable operations at the same time and enhance the reliability.

(91) In addition, the kind of unstable operations that will occur can be determined by using two or more types of sensors, two or more types of computation units, and two or more types of determination units and mounting them at suitable circumferential or axial positions in the turbo-machine.

(92) The computation unit(s), the determination unit(s), and the control unit(s) according to the present invention is executable by a computer. Those computation unit(s), determination unit(s), and control unit(s) may be considered as programs executable by a computer.

(93) The present invention can be applied to gas turbine engines for aircraft or watercraft for enhancing the safety during the operation. Moreover, the present invention can be applied to gas turbines for power generation, steam turbines, or wind turbines for power generation for monitoring the operation stability during the operation and enhancing the reliability of the electric power supply.

REFERENCE SIGNS LIST

(94) 1 observation apparatus 2 turbo-machine 2a blade 2b wall surface 3 operation control apparatus 4 reporting unit 10 detection unit 11, 12 sensor 20 computation unit 30 determination unit 40 control unit 100 observation apparatus