Sign detection device, sign detection method, and sign detection program

Abstract

The present application relates to a sign detection device for detecting a sign of vibration that would occur in a combustor of a gas turbine. This device detects a sign on the basis of a first index and a second index. The first index is calculated as an index indicating at least one of regression between a plurality of multidimensional vectors in which vector components are a plurality of pieces of data included in time series data that is acquired from a first sensor disposed in the combustor and intermittency between the plurality of multidimensional vectors. The second index is calculated as an index indicating synchronism between a plurality of pieces of time series data which are acquired from a second sensor and a third sensor that are disposed at different positions in the combustor.

Description

Sign detection device, sign detection method, and sign detection program

[0001] This application claims priority to Japanese Patent Application No. 2023-0140702, filed with the Japan Patent Office on August 31, 2023, the contents of which are incorporated herein by reference.

[0002] In gas turbines driven by combustion gases produced by burning fuel, unstable vibrations (sudden vibrations) that tend to fluctuate suddenly can occur in the combustor. These sudden vibrations reach a limit cycle shortly after the vibrations increase. When the sudden vibrations reach the limit cycle, they can cause a trip or place a heavy load on the equipment.

[0003] Therefore, it is desirable to avoid such sudden vibrations at an early stage. However, in the case of sudden vibrations, the increase in vibration until the limit cycle is reached occurs in a short period of time, and therefore, there is a risk that the sudden vibrations cannot be avoided by control after the actual detection of the increase in vibration. Therefore, in order to avoid sudden vibrations, it is necessary to detect the signs of sudden vibrations sufficiently before they occur. For example, Patent Document 1 discloses a technology for detecting the signs of combustion vibrations that occur in a coupled manner in multiple combustors of a gas turbine, based on parameters that indicate the correlation between physical quantities obtained from sensors provided in each of the multiple combustors.

[0004] Japanese Patent Application Laid-Open No. 2021-162397

[0005] The technology disclosed in Patent Document 1 targets the detection of signs of combustion oscillations having a frequency band of several hundred hertz that are generated in a coupled manner across multiple combustors, but is unable to detect signs of combustion oscillations in other frequency bands that are generated in only a single combustor. In recent years, the use of fuels with high combustion speeds, such as hydrogen gas, has been considered for use in gas turbines. Because hydrogen gas has a higher combustion speed than fuels such as natural gas, which have traditionally been used, the length of the flame formed in the combustor is shorter, making it more likely that combustion oscillations in higher frequency bands will occur.

[0006] At least one embodiment of the present disclosure has been made in consideration of the above-mentioned circumstances, and aims to provide a precursor detection device, a precursor detection method, and a precursor detection program that can effectively detect precursors of combustion oscillations with various characteristics.

[0007] To solve the above-described problems, at least one embodiment of the present disclosure provides a sign detection device for detecting a sign of vibration occurring in a combustor of a gas turbine, the sign detection device comprising: a first index calculation unit configured to calculate a first index indicating at least one of recurrence between a plurality of multidimensional vectors, the plurality of data being included in time-series data acquired from a first sensor installed in the combustor as vector components, and intermittency between the plurality of multidimensional vectors; a second index calculation unit configured to calculate a second index indicating synchronicity between a plurality of time-series data acquired from a second sensor and a third sensor installed at different positions in the combustor, respectively; and a sign detection unit configured to detect the sign based on the first index and the second index.

[0008] To solve the above-described problems, a sign detection method according to at least one embodiment of the present disclosure is a sign detection method for detecting a sign of vibration occurring in a combustor of a gas turbine, the sign detection method including: calculating a first index indicating at least one of recurrence between a plurality of multidimensional vectors, the plurality of data being included in time-series data acquired from a first sensor installed in the combustor, and intermittency between the plurality of multidimensional vectors as vector components; calculating a second index indicating synchronicity between a plurality of time-series data acquired from a second sensor and a third sensor installed at different positions in the combustor, respectively; and detecting the sign based on the first index and the second index.

[0009] To solve the above-described problems, a sign detection program according to at least one embodiment of the present disclosure is a sign detection program for detecting a sign of vibration occurring in a combustor of a gas turbine, the program being capable of causing a computer device to execute the following steps: calculating a first index indicating at least one of recurrence between a plurality of multidimensional vectors, the plurality of data being included in time-series data acquired from a first sensor installed in the combustor as vector components, and intermittency between the plurality of multidimensional vectors; calculating a second index indicating synchronicity between a plurality of time-series data acquired from a second sensor and a third sensor installed at different positions in the combustor, respectively; and detecting the sign based on the first index and the second index.

[0010] According to at least one embodiment of the present disclosure, it is possible to provide a sign detection device, a sign detection method, and a sign detection program that can suitably detect signs of combustion oscillations having various characteristics.

[0011] 11A is a schematic cross-sectional view taken along a turbine axis in the vicinity of a combustor included in a gas turbine according to an embodiment. FIG. 1 is a cross-sectional view taken along the line A-A in FIG. 1. FIG. 11B is a block diagram of a symptom detection device according to an embodiment. FIG. 11C is an example of a multidimensional vector in a D-dimensional phase space. FIG. 11D is an example of a recurrence plot corresponding to FIG. 4. FIG. 11E is a diagram showing a correspondence relationship between the distribution of regression points in the recurrence plot and the index De. FIG. 11F is a diagram showing a correspondence relationship between the distribution of regression points in the recurrence plot and the index La. FIG. 11F is an example of a distribution of regression points in a joint recurrence plot. FIG. 11G is an explanatory diagram for determining a cross-recurrence plot from the distance between data points of two pieces of time-series data in a phase space. FIG. 11H is a conceptual diagram of an order parameter K. FIG. 11H is past data of physical quantities acquired from a first sensor when combustion oscillation having a relatively low frequency band near 3000 Hz is observed. FIG. 11H is a recurrence plot corresponding to FIG. 11A. FIG. 11H is a diagram showing an example of a clustering result for N pieces of data defined by average values ​​μDe and μLa corresponding to the indexes De and La, respectively, corresponding to FIG. 11A. FIG. 10 is a diagram showing an example of the results of clustering past data of the average value μCD corresponding to the index CD, which is the second index, when combustion oscillation having a relatively high frequency band around 4500 Hz is observed.

[0012] Hereinafter, several embodiments will be described with reference to the accompanying drawings. However, the dimensions, materials, shapes, relative arrangements, etc. of the components described as embodiments or shown in the drawings are merely illustrative examples and are not intended to limit the scope of the invention.

[0013] First, a gas turbine 1 that is a detection target of a symptom detection device according to at least one embodiment of the present disclosure will be described with reference to Fig. 1. Fig. 1 is a schematic cross-sectional view along a turbine shaft in the vicinity of a combustor 8 provided in the gas turbine 1 according to one embodiment, and Fig. 2 is a cross-sectional view taken along the line A-A in Fig. 1.

[0014] The gas turbine 1 includes a compressor 2, combustors 4, stator vanes 6, and rotor blades 8. The combustors 4 are can-type combustors, and a plurality of them are arranged along the circumferential direction of the turbine shaft. The combustors 4 have the same configuration, and FIG. 1 shows only one of the combustors 4 as a representative. Each combustor 4 includes a fuel nozzle 10 and a transition piece 12.

[0015] Furthermore, at least one sensor is installed in the combustor 4 of the gas turbine 1 to detect a warning sign using a warning sign detection device 100 (described later). In FIG. 1 , three sensors (a first sensor 20a, a second sensor 20b, and a third sensor 20c) are installed in the transition piece 12 of the combustor 4. In this embodiment, for convenience of explanation, the first sensor 20a, the second sensor 20b, and the third sensor 20c are installed at different positions. However, it is sufficient that at least the second sensor 20b and the third sensor 20c are installed at different positions (i.e., the first sensor 20a may be installed at the same position as the second sensor 20b or the third sensor 20c). Furthermore, as shown in FIG. 2 , the second sensor 20b and the third sensor 20c may be arranged in the transition piece 12 of the combustor 4 along the circumferential direction with respect to the central axis or along the axial direction. Note that the second sensor 20b or the third sensor 20c may also serve as the first sensor 20a.

[0016] These sensors (first sensor 20 a, second sensor 20 b, and third sensor 20 c) are capable of measuring physical quantities at their respective positions. The physical quantities that the sensors can measure are not limited as long as they are related to the occurrence of combustion oscillation, and may be, for example, at least one of pressure, strain, acceleration, velocity, and displacement.

[0017] Next, a description will be given of a symptom detection device 100 for detecting a symptom of combustion fluctuation in the gas turbine 1 having the above configuration. The symptom detection device 100 is a computer including, for example, a CPU (Central Processing Unit), RAM (Random Access Memory), ROM (Read Only Memory), etc. In the symptom detection device 100, the processor (CPU) executes programs stored in the memory (RAM or ROM), thereby realizing various functions described below.

[0018] 3 is a block diagram of a sign detection device 100 according to an embodiment. The sign detection device 100 includes a time-series data acquisition unit 110, an index calculation unit 120, a sign detection unit 130, and an output unit 140.

[0019] The time-series data acquiring unit 110 is configured to acquire time-series data of physical quantities from each sensor (the first sensor 20 a, the second sensor 20 b, and the third sensor 20 c). The time-series data includes a plurality of data samples sampled at a plurality of timings (i.e., a predetermined sampling period) in a unit time in the most recent past.

[0020] The index calculation unit 120 is configured to calculate an index for determining whether or not there is a sign of combustion fluctuation, based on the time-series data acquired by the time-series data acquisition unit 110. In the present embodiment, the index calculation unit 120 is capable of calculating a first index I1 and a second index I2 as the indexes, and includes a first index calculation unit 122 for calculating the first index I1 and a second index calculation unit 124 for calculating the second index I2 (specific methods for calculating the first index I1 and the second index I2 will be described later).

[0021] The sign detection unit 130 is configured to detect a sign of combustion oscillation based on the index calculated by the index calculation unit 120. The sign detection unit 130 can preferably detect signs of combustion oscillation with different characteristics by using both the first index I1 calculated by the first index calculation unit 122 and the second index I2 calculated by the second index calculation unit 124 as the indexes calculated by the index calculation unit 120. That is, the combustion oscillation to be detected may have different characteristics depending on, for example, the frequency band. For example, some combustion oscillations are highly sensitive to the first index I1 and less sensitive to the second index I2, while others are less sensitive to the first index I1 and more sensitive to the second index I2. Therefore, attempting to perform sign detection based on only one of the first index I1 or the second index I2 may miss some characteristics of the combustion oscillation. Therefore, the sign detection unit 130 can preferably detect combustion oscillations with various characteristics by performing sign detection based on both the first index I1 and the second index I2.

[0022] The output unit 140 is configured to perform various outputs based on the detection results by the sign detection unit 130. The output contents by the output unit 140 may include, for example, various data indicating the signs detected by the sign detection unit 130, as well as related data such as time-series data acquired by the time-series data acquisition unit 110 and indices calculated by the index calculation unit 120. When a sign is detected by the sign detection unit 130, the output unit 140 may output a visual or audio alarm to notify the operator of the detection, or may automatically output a control signal to another control device to avoid combustion oscillations.

[0023] Next, a calculation method of the index in the index calculation unit 120 will be specifically described. First, the first index calculation unit 122 of the index calculation unit 120 calculates a first index I1 as an index. The first index I1 is calculated based on time series data acquired from the first sensor 20a installed in the combustor 4. That is, the first index I1 is an index that can be calculated from data acquired from a single sensor. The first index I1 is calculated as an index related to at least one of the recurrence between a plurality of multidimensional vectors, each of which has a plurality of data included in the time series data acquired from the first sensor 20a as vector components, and the intermittence between a plurality of multidimensional vectors. Note that the recurrence refers to the continuation over time of similarity between data at two different times in one time series signal (for example, time t ｉ and time (t ｉ +Δt), time (t ｉ＋１ ) and time (t ｉ＋１ +Δt), time (t ｉ＋２ ) and time (t ｉ＋２ +Δt)), and intermittency means that in a time series signal, similarity continues between data at a certain time and multiple times that are consecutive in time (for example, at time t ｉ and time (t ｉ +Δt), time ti and time (t ｉ＋１ +Δt), time t ｉ and time (t ｉ＋２ +Δt) (the similarity continues).

[0024] Specifically, the time series data x(t ｉ ) (i=1, 2, . . . , N) includes N pieces of measurement data acquired at predetermined sampling intervals. ｉ ) (i = 1, 2, ..., N), the delay time τ, the multidimensional vector X(t ｉ Specifically, a multidimensional vector X(t ｉ ) is defined by the following formula: ｉ ) = {x(t ｉ ), x(t ｉ+1),...,x(t ｉ +(D-1)τ)} (1)

[0025] Figure 4 shows the multidimensional vector X(t ｉ ), and FIG. 5 is an example of a recurrence plot corresponding to FIG. 4. The D-dimensional topological space is a multidimensional space in which each axis is represented by x(t), x(t+τ), x(t+2τ), ..., and FIG. 4 illustrates a case where D=3, which can be represented on paper. The first index calculation unit 122 calculates the recurrence plot of any two times (t ｉ , t ｊ ) the Euclidean distance X(t ｉ ) -X(t ｊ ) and the threshold ε, a recurrence plot is performed based on the following equation: In the recurrence plot obtained in this way, as shown in FIG. 5, the Euclidean distance X(t ｉ ) -X(t ｊ ) below a threshold ε, recurrence points with similar behavior are highlighted.

[0026] The first index calculation unit 122 can calculate, as an example of the first index I1 related to recurrence, an index De (Determinism), which is the ratio of the recurrence points constituting the diagonal line structure included in the recurrence plot to all recurrence points. FIG. 6 illustrates the correspondence between the distribution of recurrence points in the recurrence plot and the index De. In FIG. 6, the length of a representative diagonal line structure is indicated by the symbol I. When the index De is low, the number and length of the diagonal line structures included in the distribution of recurrence points in the recurrence plot are small, and the distribution of the recurrence points becomes more random. On the other hand, when the index De is high, the number and length of the diagonal line structures included in the distribution of recurrence points in the recurrence plot are large. Thus, recurrence means that similar behavior occurs repeatedly at a predetermined cycle, and can be treated as an index related to combustion oscillation. The index De related to recurrence can be quantitatively defined, for example, by the following equation: The index De defined by the above formula (3) can take a value in the range of 0≦De≦1, and the higher the periodicity, the closer to 1 it indicates a value.

[0027] Furthermore, the first index calculation unit 122 can calculate, as an example of the first index I1 related to intermittency, an index La (Laminarity), which is the ratio of the regression points constituting the vertical or horizontal structure included in the recurrence plot to the total regression points. FIG. 7 shows the correspondence between the distribution of regression points in the recurrence plot and the index La. In FIG. 7, the length of a representative vertical structure is indicated by the symbol v. When the index La is low, the number and length of vertical structures included in the distribution of regression points in the recurrence plot are small, and the distribution of regression points becomes closer to random. On the other hand, when the index La is high, the number and length of vertical structures included in the distribution of regression points in the recurrence plot are large. Thus, intermittency means that similar behavior occurs repeatedly between consecutive times and can be treated as an index related to combustion oscillation. The index La related to intermittency can be quantitatively defined, for example, by the following equation: The index La defined by the above formula (4) can take a value in the range of 0≦La≦1, and the smaller the frequency of the periodic function, the closer to 1 it indicates a value.

[0028] Subsequently, the second index calculation unit 124 of the index calculation unit 120 calculates a second index I2 as an index. The second index I2 is calculated based on the synchronicity of multiple pieces of time-series data acquired from the second sensor 20b and the third sensor 20c, which are respectively installed at different positions in the combustor 4. That is, the second index I2 is an index that can be calculated from data acquired from multiple sensors installed at different positions.

[0029] The second index calculation unit 124 can calculate, as an example of the second index I2 related to synchrony, an index De (Determinism), which is the ratio of the recurrence points constituting the diagonal line structure included in the joint recurrence plot to all recurrence points. The joint recurrence plot is an extension of the recurrence plot obtained based on the time series data acquired from the single first sensor 20a described above, in order to apply it to multiple time series data (multivariate time series data) acquired from the second sensor 20b and the third sensor 20c.

[0030] Specifically, the time series data x(t ｉ ) (i=1, 2, . . . , N), and the time series data y(t ｉ ) (i = 1, 2, ..., N) in the phase space, and if both recur simultaneously, the return points are plotted in black at the corresponding coordinates; otherwise, the return points are plotted in white. Specifically, the joint recurrence plot JR(i, j) is defined by the following equation (where i, j = 1, 2, ..., Np, where Np is the number of data points in the phase space): However, X(t ｉ ) = {x(t ｉ ), x(t ｉ +τ),...,x(t ｉ +(D-1)τ)}, Y(ti)={y(t ｉ ), y(t ｉ +τ),...,y(t ｉ +(D-1)τ)}, where D is the embedding dimension, τ is the delay time, ε is the threshold, and Θ is the Heaviside function.

[0031] 8 shows an example of the distribution of the return points in the joint recurrence plot. As shown in FIG. 8, the joint recurrence plot JR(i, j) is a distribution of the time series data X(t ｉ ) = {x(t ｉ ), x(t ｉ +τ),...,x(t ｉ+(D−1)τ)} and the time series data Y(t ｉ ) = {y(t ｉ ), y(t ｉ +τ),...,y(t ｉ +(D−1)τ)}, the common points with the recurrence plot Ry(i, j).

[0032] The second index calculation unit 124 can calculate, for such a joint recurrence plot JR(i,j), the degree of determinacy JD, which is the ratio of the recurrence points constituting the diagonal structure to all the recurrence points, as the second index I2. The index JD for the joint recurrence plot JR(i,j) is quantitatively defined by the following equation:

[0033] Furthermore, the second index calculation unit 124 can calculate an index De (Determinism), which is the ratio of the recurrence points constituting the diagonal line structure included in the cross-recurrence plot to all recurrence points, as an example of the second index I2 related to synchrony. The cross-recurrence plot is an extension of the recurrence plot obtained based on the time-series data acquired from the single first sensor 20a described above, in order to apply it to multiple time-series data (multivariate time-series data) acquired from the second sensor 20b and the third sensor 20c.

[0034] FIG. 9 is an explanatory diagram for determining a cross-recurrence plot from the distance between data points of two time-series data in a phase space. Specifically, the time-series data x(t ｉ ) (i=1, 2, . . . , N), and the time series data y(t ｉ ) (i=1, 2, ..., N) are embedded in a common topological space, and the time series data x(t ｉ ) and y(t ｉIf the distance between the points i, j is small (smaller than the threshold ε), the return points are plotted in black at the corresponding coordinates, otherwise the return points are plotted in white. Specifically, the cross recurrence plot CR(i, j) is defined by the following equation (where i, j = 1, 2, ..., Np, where Np is the number of data points in the phase space): However, X(t ｉ ) = {x(t ｉ ), x(t ｉ +τ),...,x(t ｉ +(D-1)τ)}, Y(t ｉ ) = {y(t ｉ ), y(t ｉ +τ),...,y(t ｉ +(D-1)τ)}, where D is the embedding dimension, τ is the delay time, ε is the threshold, and Θ is the Heaviside function.

[0035] In a cross-recurrence plot, when the distance between data points in phase space decreases over a long period of time, a diagonal line structure is formed. Therefore, the second index calculation unit 124 can calculate, as the second index I2, the degree of determinacy CD, which is the ratio of the recurrence points constituting the diagonal line structure to all recurrence points for such a cross-recurrence plot CR(i,j). The index CD for the cross-recurrence plot CR(i,j) is quantitatively defined by the following equation:

[0036] Furthermore, the second index calculation unit 124 can calculate an order parameter K (Kuramoto order parameter) obtained from a plurality of time-series data as an example of the second index I2 related to synchrony. The order parameter K is calculated by multiplying the time-series data x(t ｉ ) (i=1, 2, . . . , N), and the time series data y(t ｉ ) (i=1, 2, . . . , N) to the following equation: where θj(t) represents the phase of the jth oscillator at time t. The order parameter K represents the distance between the origin and the coordinates of the center of gravity of the oscillators distributed on the unit circle, and can take on a value in the range of 0≦K≦1. Figure 10 is a conceptual diagram of the order parameter K. The closer the order parameter K is to a synchronous state, the closer it is to 1.

[0037] Next, a specific description will be given of a method for determining the presence or absence of a sign in the sign detection unit 130. The sign detection unit 130 detects a sign based on both the first index I1 calculated by the first index calculation unit 122 and the second index I2 calculated by the second index calculation unit 124. As described above, the first index I1 and the second index I2 are indexes that can determine the presence or absence of a sign from different perspectives, and therefore, by performing sign detection based on both of these indexes, the sign detection unit 130 can preferably detect signs that would be missed with only one of the indexes.

[0038] The sign detection unit 122 identifies a category corresponding to the sign by applying a classification model to the past data of the first index I1 or the second index I2, and detects the sign based on whether the first index I1 calculated by the first index calculation unit 122 and the second index I2 calculated by the second index calculation unit 124 belong to the category. The classification model used by the sign detection unit 13 can be constructed, for example, by applying clustering to the past data. In this embodiment, an example is shown in which the classification model is constructed using K-means, which is a type of clustering, but the method for constructing the classification model is not limited to this.

[0039] In the K-means method, N pieces of data x ｉ (i=1,2,...,N)=[μDe(t ｉ ), μLa(t ｉ ) ], K cluster centroids mk (k=1, 2, ..., K) are assumed, where μDe is the average value calculated from multiple indices De, and μLa is the average value calculated from multiple indices La.

[0040] Next, the N pieces of data are classified into K clusters so that the objective function J defined by the following equation is minimized. Then, a new center of gravity mk is calculated for each classified data point, and this process is repeated until there is no change in the center of gravity mk, thereby identifying classifications corresponding to the K clusters. Here, rik is an indicator function and is expressed by the following equation.

[0041] FIG. 11A shows past data of physical quantities (pressure values) acquired from the first sensor 20a when combustion oscillations having a relatively low frequency band around 3000 Hz were observed, FIG. 11B is a recurrence plot corresponding to FIG. 11A, and FIG. 11C is a diagram showing an example of clustering results for N pieces of data defined by average values ​​μDe and μLa corresponding to the indicators De and La, respectively, corresponding to 11A.

[0042] In this example, by setting K = 5, the results are clustered into five categories defined by the degree of likelihood of combustion oscillation occurring. Category 1 indicates that there is no likelihood of combustion oscillation occurring, and the combustion state is stable. On the other hand, Category 5 is the category with the highest likelihood of combustion oscillation occurring, and indicates that combustion oscillation is actually occurring. Categories 2 to 4 are between Category 1 and Category 5, and indicate a state in which combustion oscillation is not actually occurring, but the likelihood of combustion oscillation occurring is increasing, in other words, they indicate signs of combustion oscillation.

[0043] At time 1, as shown in FIG. 11A, the data belongs to Class 1. At this time, as shown in FIG. 11B, the corresponding recurrence plot shows that the data exists randomly as isolated points, and the diagonal line structures indicating signs of combustion oscillation are hardly observed. At time t2, as shown in FIG. 11A, the data belongs to Class 4. At this time, as shown in FIG. 11B, the corresponding recurrence plot shows that the diagonal line structures indicating signs of combustion oscillation are increasing compared to time t1 (particularly, diagonal line structures appear partially in the area surrounded by the dashed line). At time t3, as shown in FIG. 11A, the data belongs to Class 5. At this time, as shown in FIG. 11B, the corresponding recurrence plot shows many diagonal line structures overall, indicating that combustion oscillation is actually occurring.

[0044] As described above, the results of clustering analysis of past data using the indices De and La, which are the first indices I, favorably observed the occurrence of combustion oscillations having a frequency band near 3000 Hz. On the other hand, although not shown in FIGS. 11A to 11C, the second indices I2 calculated based on the past data shown in FIG. 11A did not allow observation of changes due to the combustion oscillations. From this, it was found that the second indices I2 had poor sensitivity to combustion oscillations having a relatively low frequency band near 3000 Hz, while the first indices I1 had excellent sensitivity.

[0045] According to the inventor's verification, combustion oscillations having a frequency band near 3000 Hz exhibit not only a specific central frequency, but also a behavior in which the oscillation frequency transitions to surrounding frequencies over time (in other words, a behavior in which the oscillation mode switches rapidly between a plurality of oscillation modes having mutually close oscillation frequencies). Therefore, it appears that the second index I2 has poor sensitivity to synchronicity, and the first index I1 has excellent sensitivity to recurrence and intermittency.

[0046] From the above results, it was found that when the combustion oscillation to be detected has a relatively low frequency band, it is effective for the sign detection unit 130 to detect the presence or absence of a sign based on the first index I1. In this case, the sign detection unit 130 identifies a classification corresponding to the sign stage by performing clustering analysis on past data based on the indexes De and La, which are the first index I1, as illustrated in Figures 11A to 11C. Then, based on the time-series data acquired from the actual engine by the time-series data acquisition unit 110, it is possible to determine the presence or absence of a sign of combustion oscillation based on whether the first index I1 calculated by the first index calculation unit 122 belongs to the corresponding classification.

[0047] Next, FIG. 12 shows an example of the clustering results of past data of the average value μCD corresponding to the index CD (see Equation (8) above), which is the second index I2, when combustion oscillation having a relatively high frequency band around 4500 Hz is observed. In this example, by setting K = 3, the clustering results are shown for three categories defined by the degree of possibility of combustion oscillation. Category 1 indicates that the combustion state is stable because there is no possibility of combustion oscillation. On the other hand, Category 3 is the category with the highest possibility of combustion oscillation and indicates that combustion oscillation is actually occurring. Category 2 is between Category 1 and Category 3 and indicates a state in which combustion oscillation is not actually occurring but the possibility of combustion oscillation is increasing, that is, a category that indicates signs of combustion oscillation. Note that FIG. 12 illustrates a case in which the index CD is calculated based on pressures from the first sensor 20a, the second sensor 20b, the third sensor 20c, and other sensors 20d to 20f as multiple sensors capable of acquiring physical quantities (pressures).

[0048] As shown in FIG. 12 , at time t4, the data belongs to Classification 1. At this time, the index CD calculated from the cross recurrence plot indicates that the correlation between the time series data acquired from each sensor is small, meaning that no combustion oscillation is occurring. At time t5, the data belongs to Classification 2. At this time, the index CD calculated from the cross recurrence plot indicates that the correlation between the time series data acquired from each sensor increases, meaning that there are signs of combustion oscillation. At time t6, the data belongs to Classification 3. At this time, the index CD calculated from the cross recurrence plot indicates that the correlation between the time series data acquired from each sensor increases, meaning that combustion oscillation is actually occurring.

[0049] As described above, the results of clustering analysis of past data using the index CD, which is the second index I2, clearly show that combustion oscillations having a frequency band near 4500 Hz occur. On the other hand, although not shown in Figure 12, the first index I1 calculated based on past data did not allow observation of any changes due to the combustion oscillations. This indicates that the first index I1 has poor sensitivity to combustion oscillations having a relatively high frequency band near 4500 Hz, while the second index I2 has excellent sensitivity.

[0050] According to the inventor's verification, combustion oscillations having a frequency band near 3000 Hz exhibit not only a specific central frequency, but also a behavior in which the oscillation frequency transitions to surrounding frequencies over time (in other words, a behavior in which the oscillation mode switches rapidly between a plurality of oscillation modes having mutually close oscillation frequencies). Therefore, it appears that the second index I2 has poor sensitivity to synchronicity, and the first index I1 has excellent sensitivity to recurrence and intermittency.

[0051] From the above results, it has been found that when the combustion oscillation to be detected has a relatively high frequency band, it is effective for the sign detection unit 130 to detect the presence or absence of a sign based on the second index I2. In this case, the sign detection unit 130 identifies a category corresponding to the sign stage by performing clustering analysis on past data based on the index CD, which is the second index I2, as illustrated in FIG. 12 . Then, the presence or absence of a sign of combustion oscillation can be determined based on whether the second index I2 calculated by the second index calculation unit 124 from the time-series data acquired from the actual engine by the time-series data acquisition unit 110 belongs to the category.

[0052] As described above, the sign detection unit 130 performs sign detection based on both the first index I1 calculated based on time-series data acquired from the first sensor 20a and the second index I2 calculated based on multiple pieces of time-series data acquired from the second sensor 20b and the third sensor 20c. In the former, the first index I1, which indicates at least one of recurrence and intermittency of each piece of data included in the time-series data acquired by a single sensor, is used as an index for sign detection. In the latter, the second index I2, which indicates synchrony between multiple pieces of time-series data acquired by multiple sensors located at different positions, is used as an index for sign detection. By performing sign detection based on the first index I1 and the second index I2, which can evaluate combustion oscillation from different perspectives, combustion oscillations with various oscillation characteristics can be detected early at the sign stage.

[0053] In addition, within the scope of the present disclosure, the components in the above-described embodiments may be replaced with well-known components as appropriate, and the above-described embodiments may be combined as appropriate.

[0054] The contents described in each of the above embodiments can be understood, for example, as follows.

[0055] (1) According to one aspect, a sign detection device is provided for detecting a sign of vibration occurring in a combustor of a gas turbine, the sign detection device including: a first index calculation unit for calculating a first index indicating at least one of recurrence between a plurality of multidimensional vectors, the plurality of data being included in time-series data acquired from a first sensor installed in the combustor and intermittency between the plurality of multidimensional vectors as vector components; a second index calculation unit for calculating a second index indicating synchronicity between a plurality of time-series data acquired from a second sensor and a third sensor installed at different positions of the combustor, respectively; and a sign detection unit for detecting the sign based on the first index and the second index.

[0056] According to the above aspect (1), precursor detection is performed based on both a first index calculated based on time-series data acquired from a first sensor and a second index calculated based on multiple time-series data acquired from a second sensor and a third sensor. In the former, the first index indicating at least one of recurrence and intermittency of each piece of data included in the time-series data is used as the index for detecting the precursor, based on multiple pieces of data included in the time-series data acquired by a single sensor. In the latter, the second index indicating synchrony between multiple pieces of time-series data acquired by multiple sensors located at different positions is used as the index for detecting the precursor. By performing precursor detection based on the first index and the second index, which can evaluate combustion oscillation from different perspectives, combustion oscillations with various oscillation characteristics can be detected early, even at the precursor stage.

[0057] (2) In another aspect, in the aspect (1) above, the sign detection unit identifies a classification corresponding to the sign by applying a classification model to past data of the first index or the second index, and detects the sign based on whether the first index or the second index belongs to the classification.

[0058] According to the above aspect (2), the presence or absence of an indication can be detected based on whether or not each data item included in the time series data acquired from the sensor belongs to a classification identified by applying a classification model to past data of the first index or the second index.

[0059] (3) In another aspect, in the aspect (1) or (2) above, the first index calculation unit calculates the first index indicating the recurrence as a parameter indicating the ratio of regression points constituting a diagonal line structure included in a recurrence plot corresponding to the plurality of multidimensional vectors to all regression points.

[0060] According to the above aspect (3), a recurrence plot is created corresponding to a plurality of multidimensional vectors having a plurality of data included in the time-series data acquired from the first sensor as vector components, and a first index indicating recurrence is quantitatively calculated as a parameter (e.g., a determinism of the recurrence plot) indicating the ratio of the recurrence points constituting the diagonal line structure included in the recurrence plot to all the recurrence points.

[0061] (4) In another aspect, in any one of the above aspects (1) to (3), the first index calculation unit calculates the first index indicating the intermittency as a parameter indicating the ratio of a regression point constituting a vertical structure or a horizontal structure included in a recurrence plot corresponding to the plurality of multidimensional vectors to all regression points.

[0062] According to the above aspect (4), a recurrence plot is created corresponding to a plurality of multidimensional vectors having a plurality of data included in the time-series data acquired from the first sensor as vector components, and a first index indicating intermittency is quantitatively calculated as a parameter (e.g., laminarity of the recurrence plot) indicating the ratio of the recurrence points constituting the vertical structure or horizontal structure included in the recurrence plot to all the recurrence points.

[0063] (5) In another aspect, in any one of the above aspects (1) to (4), the second index calculation unit calculates the second index as at least one of an order parameter determined from the plurality of time series data, a parameter indicating a ratio of regression points forming a diagonal line structure included in a joint recurrence plot determined from the plurality of time series data to all regression points, or a parameter indicating a ratio of regression points forming a diagonal line structure included in a cross recurrence plot determined from the plurality of time series data to all regression points.

[0064] According to the above aspect (5), the second index is quantitatively calculated as an order parameter obtained from a plurality of time-series data acquired from the second sensor and the third sensor, and as a parameter indicating the ratio of the regression points constituting the oblique line structure included in the joint recurrence plot or the cross recurrence plot to all regression points (for example, Determinism of the joint recurrence plot or the cross recurrence plot).

[0065] (6) In another aspect, in any one of the above aspects (1) to (5), the second sensor and the third sensor are arranged along a circumferential direction with respect to a central axis of the combustor.

[0066] According to the above aspect (6), the second sensor and the third sensor for acquiring time series data used in calculating the second index are arranged along the circumferential direction with respect to the central axis of the combustor, thereby making it possible to acquire time series data at positions different from each other.

[0067] (7) A sign detection method according to one aspect is a sign detection method for detecting a sign of vibration occurring in a combustor of a gas turbine, the sign detection method comprising: a step of calculating a first index indicating at least one of recurrence between a plurality of multidimensional vectors, the plurality of data being included in time-series data acquired from a first sensor installed in the combustor and intermittency between the plurality of multidimensional vectors as vector components; a step of calculating a second index indicating synchronism between a plurality of time-series data acquired from a second sensor and a third sensor installed at different positions in the combustor, respectively; and a step of detecting the sign based on the first index and the second index.

[0068] According to the above aspect (7), precursor detection is performed based on both a first index calculated based on time-series data acquired from a first sensor and a second index calculated based on multiple time-series data acquired from a second sensor and a third sensor. In the former, the first index indicating at least one of recurrence and intermittency of each piece of data included in the time-series data acquired by a single sensor is used as the index for precursor detection. In the latter, the second index indicating synchrony between multiple pieces of time-series data acquired by multiple sensors located at different positions is used as the index for precursor detection. By performing precursor detection based on the first index and the second index, which can evaluate combustion oscillation from different perspectives, combustion oscillations with various oscillation characteristics can be detected early, even at the precursor stage.

[0069] (8) A sign detection program according to one aspect is a sign detection program for detecting a sign of vibration occurring in a combustor of a gas turbine, the program being executable by a computer device to: calculate a first index indicating at least one of recurrence between a plurality of multidimensional vectors, the plurality of data being included in time-series data acquired from a first sensor installed in the combustor as vector components, or intermittency between the plurality of multidimensional vectors; calculate a second index indicating synchronicity between a plurality of time-series data acquired from a second sensor and a third sensor installed at different positions of the combustor, respectively; and detect the sign based on the first index and the second index.

[0070] According to the above aspect (8), precursor detection is performed based on both a first index calculated based on time-series data acquired from a first sensor and a second index calculated based on multiple time-series data acquired from a second sensor and a third sensor. In the former, the first index indicating at least one of recurrence and intermittency of each piece of data included in the time-series data acquired by a single sensor is used as the index for detecting the precursor. In the latter, the second index indicating synchrony between multiple pieces of time-series data acquired by multiple sensors located at different positions is used as the index for detecting the precursor. By performing precursor detection based on the first index and the second index, which can evaluate combustion oscillation from different perspectives, combustion oscillations with various oscillation characteristics can be detected early, even at the precursor stage.

[0071] REFERENCE SIGNS LIST 1 Gas turbine 2 Compressor 4 Combustor 6 Stationary vane 8 Rotor blade 10 Fuel nozzle 12 Transition piece 20a First sensor 20b Second sensor 20c Third sensor 100 Sign detection device 110 Time series data acquisition unit 120 Index calculation unit 130 Sign detection unit 140 Output unit

Claims

1. A sign detection device for detecting a sign of vibration occurring in a combustor of a gas turbine, comprising: a first index calculation unit for calculating a first index indicating at least one of a regression between a plurality of multidimensional vectors, the plurality of data being included in time series data acquired from a first sensor installed in the combustor, and an intermittency between the plurality of multidimensional vectors, the plurality of data being vector components; a second index calculation unit for calculating a second index indicating synchronicity between a plurality of time series data acquired from a second sensor and a third sensor installed at different positions of the combustor, respectively; and a sign detection unit for detecting the sign based on the first index and the second index.

2. The sign detection device according to claim 1, wherein the sign detection unit identifies a classification corresponding to the sign by applying a classification model to past data of the first indicator or the second indicator, and detects the sign based on whether the first indicator or the second indicator belongs to the classification.

3. The symptom detection device according to claim 1 or 2, wherein the first index calculation unit calculates the first index indicating the recurrence as a parameter indicating the ratio of the recurrence points constituting the diagonal line structure included in the recurrence plot corresponding to the plurality of multidimensional vectors to all the recurrence points.

4. The symptom detection device according to claim 1 or 2, wherein the first index calculation unit calculates the first index indicating the intermittency as a parameter indicating the ratio of a regression point constituting a vertical structure or a horizontal structure included in a recurrence plot corresponding to the plurality of multidimensional vectors to all regression points.

5. The symptom detection device according to claim 1 or 2, wherein the second index calculation unit calculates the second index as at least one of an order variable calculated from the plurality of time series data, a parameter indicating a ratio of regression points forming a diagonal line structure included in a joint recurrence plot calculated from the plurality of time series data to all regression points, or a parameter indicating a ratio of regression points forming a diagonal line structure included in a cross recurrence plot calculated from the plurality of time series data to all regression points.

6. A precursor detection device as described in claim 1 or 2, wherein the second sensor and the third sensor are arranged circumferentially about a central axis of the combustor.

7. A method for detecting a sign of vibration occurring in a combustor of a gas turbine, comprising: a step of calculating a first index indicating at least one of recurrence between a plurality of multidimensional vectors, the vector components of which are a plurality of data included in time-series data acquired from a first sensor installed in the combustor, or intermittency between the plurality of multidimensional vectors; a step of calculating a second index indicating synchronicity between a plurality of time-series data acquired from a second sensor and a third sensor installed at different positions of the combustor, respectively; and a step of detecting the sign based on the first index and the second index.

8. A sign detection program for detecting signs of vibration occurring in a combustor of a gas turbine, the sign detection program being capable of executing, on a computer device, the following steps: calculating a second index indicating at least one of recurrence between a plurality of multidimensional vectors, the vector components of which are a plurality of data included in time-series data acquired from a first sensor installed in the combustor, or intermittency between the plurality of multidimensional vectors; calculating a second index indicating synchronicity between a plurality of time-series data acquired from a second sensor and a third sensor installed at different positions of the combustor, respectively; and detecting the sign based on the first index and the second index.

Images