Device monitoring method, device monitoring apparatus, and storage medium

Abstract

A device monitoring method is a monitoring method of a device using Mahalanobis distance calculated from data of a plurality of variables indicating a state of the device, wherein the method includes: a division step of dividing a range of one variable indicating the state of the device into a plurality of first range bands based on a frequency distribution of the one variable; and a unit space creation step of creating a plurality of unit spaces that become a basis for calculation of the Mahalanobis distance, respectively, from data of the plurality of variables corresponding to a plurality of second range bands of the one variable decided based on the plurality of first range bands.

Description

Technical Field

[0001] This invention relates to equipment monitoring methods, equipment monitoring devices, and equipment monitoring programs.

[0002] This application claims priority based on Japanese Patent Application No. 2021-037106, filed with the Japanese Patent Office on March 9, 2021, the contents of which are incorporated herein by reference. Background Technology

[0003] Sometimes, Mahalanobis distance, which represents the deviation between a baseline set of data representing the state of a device (such as state quantities that can be obtained by sensors) and the measured data about that variable, is used to monitor the device.

[0004] Patent Document 1 describes a device monitoring method using Mahalanobis distance, in which the Mahalanobis distance is calculated using multiple unit spaces set according to the operating period. Here, the aforementioned unit space is a collection of data that serves as a reference for determining whether the operating state of the device is normal. More specifically, in Patent Document 1, the Mahalanobis distance for data acquired during the device's start-up and operation periods is calculated using unit spaces based on the device's state quantities during the device's start-up and operation periods, and the Mahalanobis distance for data acquired during the device's load operation periods is calculated using unit spaces based on the device's state quantities during the device's load operation periods.

[0005] Existing technical documents

[0006] Patent documents

[0007] Patent Document 1: Japanese Patent No. 5031088 Summary of the Invention

[0008] The problem that the invention aims to solve

[0009] Furthermore, regarding the monitored device, the data representing the device's state is divided according to a certain benchmark, and the Mahalanobis distance is calculated using multiple unit spaces created based on each division. This is considered to improve the anomaly detection accuracy compared to the case where the Mahalanobis distance is calculated using a single unit space created using all of the aforementioned data.

[0010] However, when the data representing the state of the device is divided into multiple unit spaces as described above, the amount of data in a particular unit space may decrease depending on the method of data division, and the accuracy of detecting device anomalies may be reduced.

[0011] In view of the above, the object of at least one embodiment of the present invention is to provide a device monitoring method, device monitoring apparatus and device monitoring program capable of detecting device malfunctions with good accuracy.

[0012] Solution for solving the problem

[0013] At least one embodiment of the device monitoring method of the present invention is a device monitoring method using Mahalanobis distance calculated from data of multiple variables representing the state of the device, wherein,

[0014] The device monitoring method includes:

[0015] The partitioning step involves dividing the range of a variable representing the state of the device into multiple first range bands based on the frequency distribution of the variable; and

[0016] The unit space creation step involves creating multiple unit spaces as the basis for calculating the Mahalanobis distance, based on the data of the multiple variables corresponding to the multiple second ranges of the variable determined based on the multiple first ranges.

[0017] Furthermore, at least one embodiment of the device monitoring apparatus of the present invention is a device monitoring apparatus that uses Mahalanobis distance calculated from data of multiple variables representing the state of the device, wherein,

[0018] The equipment monitoring device includes:

[0019] A partitioning unit, configured to divide the range of a variable representing the state of the device into multiple first range bands based on the frequency distribution of the variable; and

[0020] The unit space production department produces multiple unit spaces as the basis for calculating the Mahalanobis distance, based on the data of the multiple variables corresponding to the multiple second ranges of the variable determined based on the multiple first ranges.

[0021] Furthermore, at least one embodiment of the device monitoring program of the present invention is a program for monitoring the device using Mahalanobis distance calculated from data of multiple variables representing the state of the device, wherein,

[0022] The device monitoring program causes the computer to perform the following process:

[0023] Based on the frequency distribution of a variable representing the state of the device, the range of the variable is divided into multiple first range bands; and

[0024] Based on the data of the plurality of variables corresponding to the plurality of second ranges of the variable determined based on the plurality of first ranges, a plurality of unit spaces are respectively constructed as the basis for the calculation of the Mahalanobis distance.

[0025] Invention Effects

[0026] According to at least one embodiment of the present invention, a device monitoring method, a device monitoring apparatus, and a device monitoring program are provided that are capable of detecting device malfunctions with good accuracy. Attached Figure Description

[0027] Figure 1 This is a schematic diagram of the gas turbine included in a device for applying a monitoring method according to one embodiment.

[0028] Figure 2 This is a schematic structural diagram of a steam turbine included in a device for applying a monitoring method according to one embodiment.

[0029] Figure 3 This is a schematic diagram of the device monitoring apparatus according to one embodiment.

[0030] Figure 4 This is a flowchart of a device monitoring method according to one embodiment.

[0031] Figure 5 This is a graph showing an example of the frequency distribution of a device's output (a variable).

[0032] Figure 6 This is a graph showing an example of the cumulative frequency distribution of the device's output (a variable).

[0033] Figure 7 This is a graph showing an example of the frequency distribution of a device's output (a variable).

[0034] Figure 8 This is a graph showing an example of the frequency distribution of a device's output (a variable).

[0035] Figure 9 This is a graph showing an example of the frequency distribution of a device's output (a variable).

[0036] Figure 10 It is a graph that schematically shows a portion of the frequency distribution of the device's output (a variable).

[0037] Figure 11 This is a diagram that schematically illustrates an example of a unit space. Detailed Implementation

[0038] Hereinafter, several embodiments of the present invention will be described with reference to the accompanying drawings. The dimensions, materials, shapes, and relative arrangements of the constituent components described as embodiments or shown in the drawings are not intended to limit the scope of the present invention, but are merely illustrative examples.

[0039] (Structure of the equipment monitoring device)

[0040] Figure 1 as well as Figure 2 This is a schematic diagram of the machine included in a device that applies a monitoring method according to several implementations. Figure 1 The machine shown is a gas turbine. Figure 2 The machine shown is a steam turbine. Figure 3 This is a schematic diagram of the device monitoring apparatus according to one embodiment.

[0041] Figure 1 The gas turbine 10 shown includes a compressor 12 for compressing air, a combustor 14 for burning fuel together with the compressed air from the compressor 12, and a turbine 16 driven by the combustion gases produced in the combustor 14. A generator 18 is connected to the rotor 15 of the gas turbine 10, and the generator 18 is driven by the rotation of the gas turbine 10.

[0042] Figure 2 The steam turbine 20 shown includes a boiler 22 for generating steam and a turbine 24 driven by the steam from the boiler 22. The turbine 24 includes a high-pressure turbine 25, an intermediate-pressure turbine 26 with an inlet pressure lower than that of the high-pressure turbine 25, and a low-pressure turbine 27 with an inlet pressure lower than that of the intermediate-pressure turbine 26. A reheater 29 is provided between the high-pressure turbine 25 and the intermediate-pressure turbine 26. A generator 28 is connected to the rotor 23 of the steam turbine 20, and the generator 28 is driven by the rotation of the steam turbine 20.

[0043] In several embodiments, the equipment to be monitored includes the gas turbine 10 or steam turbine 20 described above. In several embodiments, the equipment to be monitored may also include a turbine (windmill, waterwheel, etc.) driven by renewable energy sources such as wind or water power. In several embodiments, the equipment to be monitored may also include machinery other than a turbine.

[0044] Figure 3 The equipment monitoring device 40 shown is configured to monitor the equipment based on the measured values ​​of multiple variables representing the state of the equipment, which are measured by the measuring unit 30.

[0045] The measurement unit 30 is configured to measure multiple variables that indicate the state of the device. The measurement unit 30 may also include multiple sensors configured to measure multiple variables that indicate the state of the device respectively.

[0046] In the case of a device including a gas turbine 10, the measurement unit 30 may also include a sensor configured to measure any one of the following variables representing the state of the device: rotor speed of the gas turbine 10, temperature of each stage of the blade path, average temperature of the blade path, turbine inlet pressure, turbine outlet pressure, and generator output. In the case of a device including a steam turbine 20, the measurement unit 30 may also include a sensor configured to measure any one of the following variables representing the state of the steam turbine 20: rotor speed of the steam turbine 20, temperature of each stage of the blade path, average temperature of the blade path, turbine inlet pressure, turbine outlet pressure, and generator output.

[0047] The equipment monitoring device 40 is configured to receive signals from the measurement unit 30 that represent measured values ​​of variables indicating the state of the equipment. The equipment monitoring device 40 may also be configured to receive signals representing measured values ​​from the measurement unit 30 according to a predetermined sampling period. Furthermore, the equipment monitoring device 40 is configured to process the signals received from the measurement unit 30 to determine whether an equipment malfunction exists. The determination result made by the equipment monitoring device 40 may also be displayed on the display unit 60 (such as a monitor).

[0048] like Figure 3 As shown, the device monitoring device 40 in one embodiment includes a data acquisition unit 42, a division unit 44, a unit space production unit 46, a Mahalanobis distance calculation unit 48, and an anomaly determination unit 50.

[0049] The device monitoring device 40 includes a computer equipped with a processor (CPU, etc.), a storage device (memory device; RAM, etc.), an auxiliary storage unit, and an interface. The device monitoring device 40 receives signals from the measurement unit 30 via the interface, which represent measured values ​​of variables indicating the state of the device. The processor is configured to process these received signals. Furthermore, the processor is configured to process programs running in the storage device. Thus, the functions of each of the aforementioned functional units (such as the data acquisition unit 42) are realized.

[0050] The processing content in the device monitoring device 40 is installed as a program executed by the processor. The program may also be stored in an auxiliary storage unit. During program execution, these programs are expanded in the storage device. The processor reads the program from the storage device and executes the commands contained in the program.

[0051] The data acquisition unit 42 is configured to acquire data of a variable representing the state of the device at multiple times t (t1, t2, ...), and multiple variables (V1, V2, ..., Vn) representing the state of the device. In the embodiment described below, the data acquisition unit 42 is configured to acquire data of the device's output (p) as a variable representing the state of the device. It should be noted that the device's output may also be the output of a generator such as the generator 18 connected to the gas turbine 10 or the generator 28 connected to the steam turbine 20. In other embodiments, the data acquisition unit 42 may also be configured to acquire the rotational speed of the machine constituting the device, values ​​related to machine vibration (values ​​representing vibration frequency, vibration level, etc.), machine temperature, ambient temperature, or the flow rate (supply amount) of fuel supplied to the machine, etc., as a variable representing the state of the device.

[0052] The data acquisition unit 42 may also be configured to acquire the aforementioned data based on the output (one variable) or measured values ​​of multiple variables of the device measured by the measurement unit 30. The device output or the measured values ​​of multiple variables, or data based on those measured values, may also be stored in the storage unit 32. The data acquisition unit 42 may also be configured to acquire the aforementioned measured values ​​or data based on those measured values ​​from the storage unit 32.

[0053] It should be noted that the storage unit 32 may also include the main storage unit or auxiliary storage unit of the computer constituting the device monitoring device 40. Alternatively, the storage unit 32 may also include a remote storage device connected to the computer via a network.

[0054] The division unit 44 is configured to divide the output range of the device into multiple first output bands (range bands) (A1, A2, ...) based on the frequency distribution of the output (a variable) of the device obtained by the data acquisition unit 42.

[0055] The unit space production unit 46 is configured to determine multiple second output bands (range bands) (B1, B2, ...) based on multiple first output bands obtained by the division unit 44. In addition, the unit space production unit 46 is configured to produce multiple unit spaces as the basis for the calculation of Mahalanobis distance based on the data (measured values) of multiple variables (V1, V2, ..., Vn) corresponding to the multiple second output bands respectively.

[0056] The aforementioned unit space is relative to the target homogeneous group (the set of normal data). The distance of the data being evaluated from the center of the unit space is calculated using the Mahalanobis distance. A smaller Mahalanobis distance indicates a higher probability that the data being evaluated is normal, while a larger Mahalanobis distance indicates a higher probability that the data being evaluated is abnormal.

[0057] The Mahalanobis distance calculation unit 48 is configured to calculate the Mahalanobis distance for the data of the evaluation object by using the unit space corresponding to the output (one variable) of the device when acquiring the data (measured values) of multiple variables of the evaluation object in multiple unit spaces produced by the unit space production unit 46.

[0058] The anomaly determination unit 50 is configured to determine whether there is an anomaly in the device based on the Mahalanobis distance calculated by the Mahalanobis distance calculation unit 48.

[0059] (Equipment monitoring process)

[0060] The following describes several embodiments of the device monitoring method in more detail. It should be noted that the following description applies to the case where the device monitoring device 40 described above is used to perform one embodiment of the device monitoring method; however, in several embodiments, other devices may also be used to perform the device monitoring method.

[0061] Figure 4 This is a flowchart of a monitoring method for a device with several implementations. Figures 5-9 This is a diagram illustrating a monitoring method for a device according to several implementation methods. Figure 5 as well as Figures 7-9 This is a graph showing an example of the frequency distribution (frequency curve) of the device's output (a variable). Figure 6 This is a graph illustrating an example of the cumulative frequency distribution of a device's output (a variable). It should be noted that... Figure 5 as well as Figures 7-9 In the diagram, the horizontal axis represents the device's output (one variable), and the vertical axis represents the frequency of the device's output (one variable). Additionally, in... Figure 6 In the diagram, the horizontal axis represents the device's output (a variable), and the vertical axis represents the cumulative relative frequency of the device's output (a variable). Figures 7-9 In the chart, the curve representing the cumulative relative frequency is shown as a dashed line.

[0062] In several implementations, firstly, the data acquisition unit 42 acquires data of the device's output (one variable) and multiple variables representing the device's state (S2). More specifically, in step S2, the device's output p(p1, p2, ...) corresponding to each of the multiple times t (t1, t2, ...) is acquired, and data of n variables (V1, V2, ..., Vn) representing the device's state corresponding to each of the multiple times t (t1, t2, ...) are also acquired. It should be noted that the device's output corresponding to time t or the data of the aforementioned multiple variables can also be representative values ​​(e.g., average values) of the device's output or the measured values ​​of the multiple variables over a specified period based on time t.

[0063] The n variables representing the state of the equipment may include, for example, at least one of the following: the rotor speed of the gas turbine 10 or the steam turbine 20, the temperature of each stage of the blade path, the average temperature of the blade path, the turbine inlet pressure, the turbine outlet pressure, and the generator output.

[0064] Next, the division unit 44 divides the output range of the device into multiple first output bands (range bands) (A1, A2, ...) (S4) based on the frequency distribution of the device's output. The frequency distribution of the device's output can be obtained based on the device's output obtained in step S2.

[0065] Figure 5 This is a graph showing an example of the frequency distribution of the device output p obtained from step S2. Figure 5 The chart shown illustrates the frequency distribution of the device's output range, which is above 0 [MW] and below Pmax [MW].

[0066] In step S4, for example, the range of each of the multiple first output bands (A1, A2, ...) is determined in such a way that the deviation of the frequency of the output included in each of the multiple first output bands (A1, A2, ...) is not large.

[0067] Here, Figure 6 It means to Figure 5 The diagram shows a frequency distribution obtained by transforming the frequency distribution of the device's output into a cumulative frequency distribution. In several embodiments, in step S4, the range of each of the first output bands can also be determined based on the relative cumulative frequency of the device's outputs in such a way that the relative frequencies of the outputs associated with the plurality of first output bands (A1, A2...) are approximately equal (i.e., in such a way that the frequencies of the outputs associated with the plurality of first output bands are approximately equal).

[0068] When using Figure 6 When illustrating this process with a diagram, firstly, the cumulative relative frequency at output 0 is set to 0%, and the cumulative relative frequency at output Pmax is set to 100%. The cumulative relative frequency is then divided into multiple ranges: above 0% and below C1, above C1 and below C2, above C3 and below C4, above and below C4 and below C5, above C5 and below C6, and above C6 and below C7 (=100%). The relative frequency widths of these multiple ranges are approximately the same (i.e., the frequencies within the multiple ranges are approximately the same). Furthermore, the output bands corresponding to these multiple ranges can be determined as multiple first output bands (A1 to A7). Here, the output [MW] ranges of the first output bands A1 to A7 are respectively above 0% and below Pmax. A1 Below, exceeding P A1 And P A2 Below, exceeding P A2 And P A3Below, exceeding P A3 And P A4 Below, exceeding P A4 And P A5 Below, exceeding P A5 And P A6 Below, exceeding P A6 And P A7 The following is an explanation of the frequency ratios of the first output bands A1 to A7, which are represented by C1, (C2-C1), (C3-C2), (C4-C3), (C5-C4), (C6-C5), and (C7-C6), respectively.

[0069] In several embodiments, in step S4, the output range of the device is divided such that the ratio of the frequencies of the outputs of devices within any two output bands of the plurality of first output bands (A1, A2, ...) is greater than or equal to 0.75 and less than 1.25. It should be noted that when using... Figure 6 In the example shown, the ratio of the frequencies of the outputs of devices in the first output band A2 to those in the first output band A3 can be represented by (C3 - C2) / (C2 - C1).

[0070] In several implementations, in step S4, the output range of the device is divided such that the ratio of the frequencies of the outputs of the device in at least two of the plurality of first output bands (A1, A2, ...) is 1.

[0071] In several implementations, in step S4, the output range of the device is divided such that the ratio of the frequencies of the outputs of the device in any two of the plurality of first output bands (A1, A2, ...) is 1.

[0072] The following describes step S4 as follows: Figure 6 The description is based on the premise that the output range of the device is divided into 7 first output bands (A1 to A7).

[0073] Next, the unit space production unit 46 determines the multiple second output bands (range bands) (B1, B2, ...) of the device based on the multiple first output bands (A1 to A7) (S6). Here, the multiple first output bands (A1 to A7) are set based on the frequency distribution of the device's output, so the multiple second output bands (B1, B2, ...) can also be said to be determined based on the frequency distribution of the device's output. It should be noted that the process of step S6 will be described later.

[0074] Next, the unit space production unit 46 produces multiple unit spaces (Q1, Q2, ...) (S8) based on the data of n variables (multiple variables) (V1, V2, ..., Vn) corresponding to the multiple second output bands (B1, B2, ...) determined by step S6, respectively.

[0075] Furthermore, the Mahalanobis distance calculation unit 48 uses the unit space corresponding to the device output (one variable) at the time of acquisition of the data of the n variables (multiple variables) of the evaluation object from the multiple unit spaces (Q1, Q2, ...) created by the unit space creation unit 46, to calculate the Mahalanobis distance (S10) for the data (signal space data) of the evaluation object. For example, if the device output at the time of acquisition of the data of the n variables of the evaluation object is within the range of the second output band B2, the Mahalanobis distance D for the data of the evaluation object is calculated using the unit space Q2 corresponding to the second output band B2.

[0076] The Mahalanobis distance for the data of the evaluation object can be calculated using the method described in Patent Document 1, but the method for calculating the Mahalanobis distance can be summarized as follows. First, data (X1, X2, ..., Xn) related to the data constituting the unit space (n variables (V1, V2, ..., Vn)) are used. n The average of each item (variable) is calculated using the following formula (A). It should be noted that in the following formula, k is the number of data points (data sets) of each of the n variables constituting the unit space.

[0077] [Mathematical Formula 1]

[0078]

[0079] Next, using the average of each item (variable) calculated by equation (A) above, the covariance matrix COV (n×n rows and columns) is obtained for the data constituting the unit space by equation (B) below.

[0080] [Mathematical Formula 2]

[0081]

[0082] Then, using the data Y1~Y of the evaluation object. n The squared value of the Mahalanobis distance D is calculated using the average obtained through equation (A) above and the inverse of the covariance matrix obtained through equation (B) above, and then calculated using equation (C). 2 It should be noted that in the following formula, l represents the data (signal spatial data) Y1 to Y2 related to the evaluation object and n variables. n Number of data sets (number of data groups).

[0083] [Mathematical Formula 3]

[0084]

[0085] Next, the anomaly determination unit 50 determines whether or not the device is abnormal based on the Mahalanobis distance D calculated in step S10 (S12). In step S12, the determination of whether or not the device is abnormal may also be based on a comparison between the Mahalanobis distance D and a threshold. For example, it may be determined that the device is normal when the Mahalanobis distance D calculated in step S10 is below the threshold, and that the device is abnormal when the Mahalanobis distance D is greater than the threshold.

[0086] According to the method of the above-described embodiment, the output range of the device is divided into multiple first output bands (A1, A2, ...) based on the frequency distribution of the device's output, and multiple unit spaces (Q1, Q2, ...) are created corresponding to multiple second output bands (B1, B2, ...) determined based on the multiple first output bands. That is, multiple output bands (first output bands and second output bands) corresponding to multiple unit spaces are determined based on the frequency distribution of the device's output. Therefore, for example, by determining the multiple output bands (first output bands or second output bands) in a way that makes the frequencies within the multiple output bands equal, it is easy to ensure that the number of data for the multiple variables (V1, V2, ..., Vn) constituting the multiple unit spaces is sufficient. Alternatively, it is easy to avoid the situation where the number of data for any unit space constituting the multiple unit spaces is too small. Therefore, it is possible to perform device anomaly detection with good accuracy based on Mahalanobis distance, regardless of the device's output, for example, to suppress false detections and false alarms.

[0087] Furthermore, in the above embodiment, in step S4, when the output range of the device is divided such that the ratio of the frequencies within any two output bands of the plurality of first output bands (A1, A2, ...) is 0.75 or higher and 1.25 or lower, the frequencies of the outputs within each of the plurality of first output bands are approximately equal. Therefore, it is easy to ensure that the number of data constituting the plurality of variables in the plurality of unit spaces determined based on the plurality of first output bands is sufficient. Thus, it is possible to perform device anomaly detection with good accuracy based on Mahalanobis distance, independent of the device's output.

[0088] Furthermore, in the above embodiment, in step S4, when the output range of the device is divided such that the ratio of the frequencies in at least two of the plurality of first output bands (A1, A2, ...) is 1, the frequencies of the outputs in at least two of the plurality of first output bands are equal. Therefore, it is easy to ensure that the number of data constituting the plurality of variables in the plurality of unit spaces determined based on the two output bands is sufficient. Thus, it is possible to perform device anomaly detection with good accuracy based on Mahalanobis distance, independent of the device's output.

[0089] In several embodiments, in step S6, the unit space production unit 46 determines the plurality of output bands corresponding to the plurality of first output bands (A1 to A7) as the plurality of second output bands (B1 to B7) of the device. That is, as Figure 7 As shown, the output range of the multiple second output bands (B1 to B7) is equal to the output range of the multiple first output bands (A1 to A7).

[0090] According to the above implementation method, multiple second output bands (B1 to B7) can be determined as output bands corresponding to multiple first output bands (A1 to A7) in a simple process. Therefore, it is possible to perform device anomaly detection with good accuracy based on Mahalanobis distance in a simpler process, without relying on the device output.

[0091] In several embodiments, in step S6, the unit space production unit 46 selects from the plurality of first output bands (A1 to A7) the outputs that form the boundaries between the plurality of second output bands (B1, B2, ...), and determines the plurality of output bands divided by the boundaries as the plurality of second output bands.

[0092] In several implementations, such as Figure 8 as well as Figure 9 As shown, at least one of the modes Pm1 to Pm7 of the outputs within each of the multiple first output bands (A1 to A7) can be selected as the boundary between the multiple second output bands. It should be noted that... Figure 8 In the example shown, the modes Pm1 to Pm7 of the outputs within each of the multiple first output bands (A1 to A7) are selected as the boundaries between the multiple second output bands. Furthermore, the multiple second output bands (B1 to B8) are determined by dividing the output range of the device (above 0 and below Pmax) using these modes Pm1 to Pm7.

[0093] According to the above implementation, at least one of the modes (Pm1 to Pm7) of the outputs within the plurality of first output bands (A1 to A7) is used as the boundary between the plurality of second output bands (B1, B2, ...). Therefore, in the output pair frequency graph ( Figure 8 , Figure 9 In the second output band (an adjacent pair of second output bands) with the boundary as the upper or lower limit, each band contains at least approximately half of the peak area encompassing the boundary. Therefore, it is easier to ensure the number of data points that constitute the unit space (Q1, Q2, ...) corresponding to these second output bands (B1, B2, ...). Thus, the accuracy of anomaly detection in Mahalanobis distance-based devices can be improved.

[0094] Here, Figure 10It is a graph that schematically illustrates a portion of the frequency distribution of the device's output. Figure 11 It is an illustrative representation based on Figure 10 The diagram shows an example of a unit space plotted based on the frequency distribution of the device's output. Here, Figure 10 The output band B in k And B k+1 The output band is divided by the mode Pma and Pmb of the device's output, and the output band B is... j It is the output band divided by the modes of the device's output, Pc and Pd. It should be noted that... Figure 11 The ellipses in the diagram represent the unit space (Q) k Q k+1 Q j Each ellipse is a set of points whose Mahalanobis distances are equal, calculated from each unit space.

[0095] Output with B j The output is divided by the output modes (Pma, Pmb, etc.) from each other by the output Pc and Pd. Therefore, in the output band B j The data within does not contain much information related to the output band B. j The data corresponding to the output near the lower bound (Pc) and upper bound (Pd) is not simply the data near the mode Pma located between the lower and upper bounds. This means that, in representing the output band B... j The unit space Q composed of data within it j In an ellipse, there are fewer data points near the ends of the major axis, and more data points near the center of the major axis (see reference). Figure 11 In this case, the shape of the ellipse (the inclination of its major axis, etc.) is not stably determined (see reference). Figure 11 Q in j And Q j Therefore, anomaly detection based on Mahalanobis distance is unstable.

[0096] For example, the data (signal spatial data) of the evaluation object is represented as Figure 11 In the case of d in the diagram, based on unit space Q j The calculated Mahalanobis distance and the distance based on unit space Q j The calculated Mahalanobis distances differ significantly. That is, based on the unit space Q... j The calculated Mahalanobis distance is relatively large, based on Q per unit space. j The calculated Mahalanobis distance is relatively small. Therefore, there is a possibility that the anomaly detection results based on the Mahalanobis distance may differ. Consequently, for example, the possibility of misjudgment in anomaly detection increases.

[0097] On the other hand, the output band Bk The output is divided by the mode Pma and Pmb. Therefore, in the output band B... k The data within contains information related to the output band B. k The outputs near the lower limit (Pma) and upper limit (Pmb) correspond to a relatively large amount of data. This means that, in representing the output band B... k The unit space Q composed of data within it k In an ellipse, there is a greater amount of data located near the two ends of the major axis (see reference). Figure 11 In this case, the shape of the ellipse (such as the inclination of the major axis) is stably determined. Therefore, the Mahalanobis distance is stably calculated, enabling stable anomaly detection.

[0098] Additionally, regarding the representation of output band B... k Adjacent output band B k+1 The unit space Q composed of data within it k+1 Similarly, the shape of the ellipse (the inclination of its major axis, etc.) is stably determined, and the two ellipses are smoothly connected (e.g., the inclinations of these ellipses become similar). Therefore, even during equipment operation, the output of the equipment crosses the output band B. k With output band B k+1 boundary ( Figure 10 Even with changes in Pmb (in the Pmb), the anomaly detection can still be stabilized.

[0099] Regarding this point, according to the above-described implementation, the modes Pm1 to Pm7 of the outputs within the first output bands (A1 to A7) are set as the boundaries between the multiple second output bands (B1, B2, ...). Therefore, the data within the second output bands with these boundaries as upper or lower limits contains a relatively large amount of data corresponding to the outputs near these boundaries (upper or lower limits). Consequently, the connection between the unit spaces (Q1, Q2, ...) created based on the data within these second output bands (B1, B2, ...) becomes smoother. Therefore, even if the device output changes across the aforementioned boundaries, device anomalies can be detected reliably.

[0100] In several implementations, in step S6, when the difference between the modes of adjacent pairs of outputs is less than a predetermined value, the pair with the larger frequency of the mode is selected as the boundary between the second output bands, and the pair with the smaller frequency is not selected as the boundary between the second output bands.

[0101] For example, in Figure 9In the example shown, the difference between adjacent pairs of modes Pm4 and Pm5 among the modes Pm1 to Pm7 of the outputs within each of the multiple first output bands (A1 to A7) is small, less than a specified value. Therefore, the mode Pm4, which has the higher frequency among the modes Pm4 and Pm5, is selected as the boundary between the second output bands, while the mode Pm5, which has the lower frequency, is not selected as the boundary between the second output bands. As a result, the output range of the device (above 0 and below Pmax) is divided by using modes other than Pm5 among the modes Pm1 to Pm7 (i.e., Pm1 to Pm4 and Pm6 to Pm7), thereby determining the multiple second output bands (B1 to B7).

[0102] The frequency of a device's output, which becomes the peak output, sometimes varies depending on seasonal changes, etc. In such cases, it appears as different peaks located nearby in a frequency distribution graph. When the data corresponding to the outputs of such multiple peaks are contained in different unit spaces, it becomes difficult to reliably perform anomaly detection based on Mahalanobis distance. Regarding this point, according to the above-described implementation, when the difference between adjacent pairs of modes (Pm4, Pm5) of the modes (Pm1 to Pm7) of the outputs within each of the multiple first output bands (A1 to A7) is less than a predetermined value (i.e., when the aforementioned peaks are close to each other), only the one with the larger frequency (Pm4) of this pair of modes is selected as the boundary between the multiple second output bands (B1, B2, ...). Therefore, the data corresponding to these two modes (Pm4, Pm5) can be contained in the same unit space, thus enabling stable anomaly detection of the device.

[0103] The contents described in the above embodiments are as follows, for example.

[0104] (1) The device monitoring method of at least one embodiment of the present invention is a device monitoring method using Mahalanobis distance calculated from data of multiple variables representing the state of the device, wherein,

[0105] The device monitoring method includes:

[0106] The partitioning step (S4) involves dividing the range of a variable representing the state of the device (e.g., the device's output) into multiple first range bands (e.g., the aforementioned multiple first output bands A1, A2, ...); and

[0107] The unit space creation step (S6-S8) involves creating multiple unit spaces as the basis for calculating the Mahalanobis distance, based on the data of the multiple variables corresponding to the multiple second range bands (e.g., the multiple second output bands B1, B2, ... mentioned above) of the variable determined based on the multiple first range bands.

[0108] According to the method described in (1) above, the range of a variable representing the state of a device is divided into multiple first range bands based on the frequency distribution of that variable, and multiple unit spaces are created corresponding to multiple second range bands determined based on those first range bands. That is, multiple range bands (first range bands and second range bands) corresponding to the multiple unit spaces are determined based on the frequency distribution of that variable. Therefore, for example, by determining the multiple range bands (first range bands or second range bands) in a way that makes the frequencies within the multiple range bands equal, it is easy to ensure that the number of data for the multiple variables that constitute the multiple unit spaces is sufficient. Thus, it is possible to perform device anomaly detection with good accuracy based on Mahalanobis distance without relying on the value of a single variable.

[0109] (2) In several embodiments, based on the method described in (1) above,

[0110] In the division step, the range of the variable is divided such that the ratio of the frequencies of the variable in any two ranges of the plurality of first ranges is greater than 0.75 and less than 1.25.

[0111] According to the method described in (2) above, the range of a variable is divided such that the ratio of the frequencies in any two ranges among the multiple first ranges is greater than 0.75 and less than 1.25. That is, the frequencies of a variable in each of the multiple first ranges are approximately equal, thus it is easy to ensure that the number of data for multiple variables constituting multiple unit spaces determined based on the multiple first ranges is sufficient. Therefore, it is possible to perform device anomaly detection with good accuracy based on Mahalanobis distance without relying on the value of a single variable.

[0112] (3) In several embodiments, based on the method described in (1) or (2) above,

[0113] In the partitioning step, the range of the variable is partitioned such that the ratio of the frequencies of the variable in at least two of the plurality of first range bands is 1.

[0114] According to the method described in (3) above, the range of a variable is divided such that the ratio of the frequencies in at least two of the multiple first range bands is 1. That is, the frequencies of a variable in at least two of the multiple first range bands are equal, so it is easy to ensure that the number of data for multiple variables that constitute multiple unit spaces determined based on the two range bands is sufficient. Therefore, it is possible to perform device anomaly detection with good accuracy based on Mahalanobis distance without relying on the value of a single variable.

[0115] (4) In several embodiments, based on any one of the methods (1) to (3) above,

[0116] The plurality of second range bands correspond to the plurality of first range bands respectively.

[0117] According to the method described in (4) above, multiple second range zones can be determined as range zones corresponding to multiple first range zones in a simple process. Therefore, it is possible to perform equipment anomaly detection with good accuracy based on Mahalanobis distance in a simpler process, without relying on the value of a single variable.

[0118] (5) In several embodiments, based on any one of the methods (1) to (3) above,

[0119] The device monitoring method includes a boundary selection step of selecting the value of a variable from the plurality of first range bands that serves as the boundary between the plurality of second range bands.

[0120] According to the method described in (5) above, the boundaries of multiple second range zones are selected from multiple first range zones. Therefore, compared with the case where the boundaries of multiple first range zones are used as the boundaries of multiple second range zones as is, it is possible to set boundaries more suitable for creating multiple unit spaces based on the frequency distribution of a variable. As a result, the accuracy of anomaly detection in devices based on Mahalanobis distance can be improved.

[0121] (6) In several embodiments, based on the method described in (5) above,

[0122] In the boundary selection step, at least one of the modes of the variable within each of the plurality of first range bands (e.g., the modes Pm1, Pm2, ... of the output mentioned above) is selected as the boundary between the plurality of second range bands.

[0123] According to the method described in (6) above, at least one of the modes of a variable within a plurality of first range bands is used as the boundary between the plurality of second range bands. Therefore, in a graph of the frequency of a variable (e.g., output), each of the second range bands (adjacent pairs of second range bands) with the boundary as the upper or lower limit contains at least approximately half the area of ​​the peak encompassing that boundary. Thus, it is easier to ensure the number of data points that constitute the unit space corresponding to these second range bands. Therefore, the accuracy of anomaly detection in Mahalanobis distance-based devices can be improved.

[0124] Furthermore, according to the method described in (6) above, the mode of a variable within the first range is set as the boundary between multiple second ranges. Therefore, the data within the second range, with the boundary as the upper or lower limit, contains a relatively large amount of data corresponding to the value of a variable near that boundary (upper or lower limit). Thus, the connection between unit spaces created based on the data within these second ranges becomes smoother. Consequently, even if a variable changes across the aforementioned boundaries, device anomalies can be detected reliably.

[0125] (7) In several embodiments, based on the method described in (6) above,

[0126] In the boundary selection step, when the difference between the modes of two adjacent pairs of variables is less than a predetermined value, the pair with the larger frequency of the mode of the variable is selected as the boundary, and the pair with the smaller frequency is not selected as the boundary.

[0127] The value of a variable whose frequency becomes a peak sometimes varies depending on seasonal changes, etc. In this case, it appears as different peaks located nearby in the frequency distribution graph. When the data corresponding to a variable with multiple such peaks are contained in different unit spaces, it is difficult to reliably perform anomaly detection based on Mahalanobis distance. Regarding this point, according to the method described in (7) above, when the difference between adjacent pairs of modes of a variable within each of the multiple first range bands is less than a specified value (i.e., when the aforementioned peaks are close to each other), only the one with the larger frequency in this pair of modes is selected as the boundary between the multiple second range bands. Therefore, the data corresponding to these two modes can be contained in the same unit space, thus enabling stable anomaly detection of the device.

[0128] (8) In several embodiments, based on any one of the methods (1) to (7) above,

[0129] The equipment includes a gas turbine (10) or a steam turbine (20).

[0130] The variable representing the state of the device is the output of the device.

[0131] The output of the device includes the output of a generator (18, 28) connected to the gas turbine or the steam turbine.

[0132] According to the method described in (8) above, multiple range bands (a first output band and a second output band) corresponding to multiple unit spaces are determined based on the frequency distribution of the output of the generator connected to the gas turbine or steam turbine. Therefore, it is easy to ensure that the number of data for the multiple variables constituting the multiple unit spaces is sufficient. Thus, for equipment containing a gas turbine or steam turbine, anomaly detection can be performed with good accuracy based on Mahalanobis distance, independent of the equipment's output.

[0133] (9) At least one embodiment of the device monitoring device (40) is a device monitoring device that uses Mahalanobis distance calculated from data of multiple variables representing the state of the device, wherein,

[0134] The equipment monitoring device includes:

[0135] The division unit (44) is configured to divide the range of a variable representing the state of the device into multiple first range bands based on the frequency distribution of the variable; and

[0136] The unit space production unit (46) produces multiple unit spaces as the basis for calculating the Mahalanobis distance, based on the data of the multiple variables corresponding to the multiple second ranges of the variable determined based on the multiple first ranges.

[0137] According to the structure described in (9) above, the range of a variable representing the state of the device is divided into multiple first range bands based on the frequency distribution of that variable, and multiple unit spaces are created corresponding to multiple second range bands determined based on those first range bands. That is, multiple range bands (first range bands and second range bands) corresponding to the multiple unit spaces are determined based on the frequency distribution of that variable. Therefore, for example, by determining the multiple range bands (first range bands or second range bands) in a way that makes the frequencies within the multiple range bands equal, it is easy to ensure that the number of data for the multiple variables that constitute the multiple unit spaces is sufficient. Thus, it is possible to perform device anomaly detection with good accuracy based on Mahalanobis distance without relying on the value of a single variable.

[0138] (10) At least one embodiment of the device monitoring program is a program for monitoring the device using Mahalanobis distance calculated from data of multiple variables representing the state of the device, wherein,

[0139] The device monitoring program causes the computer to perform the following process:

[0140] Based on the frequency distribution of a variable representing the state of the device, the range of the variable is divided into multiple first range bands; and

[0141] Based on the data of the plurality of variables corresponding to the plurality of second ranges of the variable determined based on the plurality of first ranges, a plurality of unit spaces are respectively constructed as the basis for the calculation of the Mahalanobis distance.

[0142] According to the procedure described in (10) above, the range of a variable representing the state of the device is divided into multiple first range bands based on the frequency distribution of that variable, and multiple unit spaces are created corresponding to multiple second range bands determined based on those first range bands. That is, multiple range bands (first range bands and second range bands) corresponding to the multiple unit spaces are determined based on the frequency distribution of that variable. Therefore, for example, by determining the multiple range bands (first range bands or second range bands) in a way that makes the frequencies within the multiple range bands equal, it is easy to ensure that the number of data for the multiple variables that constitute the multiple unit spaces is sufficient. Thus, it is possible to perform device anomaly detection with good accuracy based on Mahalanobis distance without relying on the value of a single variable.

[0143] The embodiments of the present invention have been described above, but the present invention is not limited to the embodiments described above, and also includes modifications to the embodiments described above, and embodiments obtained by appropriately combining these embodiments.

[0144] In this specification, terms such as "in a certain direction," "along a certain direction," "parallel," "orthogonal," "center," "concentric," or "coaxial" indicate not only a strict configuration, but also a state of relative displacement by angle or distance with tolerance or to the extent that the same function can be obtained.

[0145] For example, terms like "same," "equal," and "homogeneous" indicate that things are equal not only in a strictly equal state, but also in a state where there is a tolerance or a difference in the degree to which they can achieve the same function.

[0146] Furthermore, in this specification, the terms "quadrilateral shape," "cylindrical shape," etc., not only refer to quadrilateral shapes and cylindrical shapes in a strict geometric sense, but also to shapes including concave and convex parts, chamfered parts, etc., within the range where the same effect can be obtained.

[0147] Furthermore, in this specification, expressions such as "possessing," "including," or "having" a constituent element are not expressions of exclusivity that exclude the existence of other constituent elements.

[0148] Explanation of reference numerals in the attached figures

[0149] 10 Gas turbine

[0150] 12 Compressors

[0151] 14. Burner

[0152] 15 Rotors

[0153] 16 Turbo

[0154] 18 Generators

[0155] 20 Steam turbines

[0156] 22 Boilers

[0157] 23 Rotors

[0158] 24 Turbo

[0159] 25 High-pressure turbine

[0160] 26 Medium-pressure turbine

[0161] 27 Low-pressure turbine

[0162] 28 Generator

[0163] 29 Reheater

[0164] 30 Measurement Department

[0165] 32 Storage Unit

[0166] 40 Equipment monitoring devices

[0167] 42 Data Acquisition Department

[0168] 44 Division

[0169] Unit 46 Space Production Department

[0170] 48. Calculation of Mahalanobis distance

[0171] 50. Anomaly Detection Department

[0172] 60 Display Section

[0173] A1~A7 First Output Band

[0174] B1~B8 Second output band.

Claims

1. A device monitoring method, which uses a Mahalanobis distance calculated from data of multiple variables representing the state of the device, wherein... The device monitoring method includes: The partitioning step involves dividing the range of a variable representing the state of the device into multiple first range bands based on the frequency distribution of the variable; and The unit space creation step involves creating multiple unit spaces as the basis for calculating the Mahalanobis distance, based on the data of the multiple variables corresponding to the multiple second ranges of the variable determined by the multiple first ranges. In the division step, the range of the variable is divided such that the ratio of the frequencies of the variable in any two ranges among the plurality of first ranges is greater than 0.75 and less than 1.

25.

2. The equipment monitoring method according to claim 1, wherein, The plurality of second range bands correspond to the plurality of first range bands respectively.

3. The equipment monitoring method according to claim 1, wherein, The equipment includes a gas turbine or a steam turbine. The variable representing the state of the device is the output of the device. The output of the device includes the output of a generator connected to the gas turbine or the steam turbine.

4. A device monitoring method, which uses a Mahalanobis distance calculated from data of multiple variables representing the state of the device, wherein... The device monitoring method includes: The partitioning step involves dividing the range of a variable representing the state of the device into multiple first range bands based on the frequency distribution of the variable; and The unit space creation step involves creating multiple unit spaces as the basis for calculating the Mahalanobis distance, based on the data of the multiple variables corresponding to the multiple second ranges of the variable determined by the multiple first ranges. In the partitioning step, the range of the variable is partitioned such that the ratio of the frequencies of the variable in at least two of the plurality of first range bands is 1.

5. The equipment monitoring method according to claim 4, wherein, The plurality of second range bands correspond to the plurality of first range bands respectively.

6. The equipment monitoring method according to claim 4, wherein, The equipment includes a gas turbine or a steam turbine. The variable representing the state of the device is the output of the device. The output of the device includes the output of a generator connected to the gas turbine or the steam turbine.

7. A device monitoring method, which uses a Mahalanobis distance calculated from data of multiple variables representing the state of the device, wherein... The device monitoring method includes: The partitioning step involves dividing the range of a variable representing the state of the device into multiple first range bands based on the frequency distribution of the variable. The unit space creation step involves creating multiple unit spaces, which serve as the basis for calculating the Mahalanobis distance, based on data corresponding to multiple second ranges of the variable determined by the multiple first ranges; and The boundary selection step involves selecting the value of a variable from the plurality of first range bands that serves as the boundary between the plurality of second range bands.

8. The equipment monitoring method according to claim 7, wherein, In the boundary selection step, at least one of the modes of the variable within each of the plurality of first range bands is selected as the boundary between the plurality of second range bands.

9. The equipment monitoring method according to claim 8, wherein, In the boundary selection step, when the difference between the modes of an adjacent pair of variables is less than a predetermined value, the mode with the larger frequency in the pair of variables is selected as the boundary, and the mode with the smaller frequency is not selected as the boundary.

10. The equipment monitoring method according to any one of claims 7 to 9, wherein, The equipment includes a gas turbine or a steam turbine. The variable representing the state of the device is the output of the device. The output of the device includes the output of a generator connected to the gas turbine or the steam turbine.

11. A device monitoring apparatus that uses a Mahalanobis distance calculated from data of multiple variables representing the state of the device, wherein, The equipment monitoring device includes: The division section is configured to divide the range of a variable representing the state of the device into multiple first range bands based on the frequency distribution of the variable. as well as The unit space creation department creates multiple unit spaces as the basis for calculating the Mahalanobis distance, based on data of multiple variables corresponding to multiple second ranges of the variable determined based on the multiple first ranges. The division section divides the range of the variable such that the ratio of the frequencies of the variable in any two ranges among the plurality of first ranges is greater than 0.75 and less than 1.

25.

12. A device monitoring apparatus that uses a Mahalanobis distance calculated from data of multiple variables representing the state of the device, wherein... The equipment monitoring device includes: The division section is configured to divide the range of a variable representing the state of the device into multiple first range bands based on the frequency distribution of the variable. as well as The unit space creation department creates multiple unit spaces as the basis for calculating the Mahalanobis distance, based on data of multiple variables corresponding to multiple second ranges of the variable determined based on the multiple first ranges. The division portion divides the range of the variable in such a way that the ratio of the frequencies of the variable in at least two of the plurality of first range bands is 1.

13. A device monitoring apparatus that uses a Mahalanobis distance calculated from data of multiple variables representing the state of the device, wherein... The equipment monitoring device includes: The division section is configured to divide the range of a variable representing the state of the device into multiple first range bands based on the frequency distribution of the variable. The unit space production department produces multiple unit spaces as the basis for calculating the Mahalanobis distance, based on the data of the multiple variables corresponding to the multiple second ranges of the variable determined based on the multiple first ranges. as well as The boundary selection unit is configured to select the value of a variable from the plurality of first range bands that serves as the boundary between the plurality of second range bands.

14. A storage medium storing a device monitoring program, the device monitoring program being a program for monitoring the device using Mahalanobis distance calculated from data of multiple variables representing the state of the device, wherein, The device monitoring program is used to cause the computer to perform the following process: Based on the frequency distribution of a variable representing the state of the device, the range of the variable is divided into multiple first range bands; and Based on the data of the plurality of variables corresponding to the plurality of second ranges of the variable determined based on the plurality of first ranges, a plurality of unit spaces are respectively constructed as the basis for the calculation of the Mahalanobis distance. In the division, the range of the variable is divided such that the ratio of the frequencies of the variable in any two ranges of the plurality of first ranges is greater than 0.75 and less than 1.

25.

15. A storage medium storing a device monitoring program, the device monitoring program being a program for monitoring the device using Mahalanobis distance calculated from data of multiple variables representing the state of the device, wherein, The device monitoring program is used to cause the computer to perform the following process: Based on the frequency distribution of a variable representing the state of the device, the range of the variable is divided into multiple first range bands; and Based on the data of the plurality of variables corresponding to the plurality of second ranges of the variable determined based on the plurality of first ranges, a plurality of unit spaces are respectively constructed as the basis for the calculation of the Mahalanobis distance. In the division, the range of the variable is divided such that the ratio of the frequencies of the variable in at least two of the plurality of first range bands is 1.

16. A storage medium storing a device monitoring program, the device monitoring program being a program for monitoring the device using Mahalanobis distance calculated from data of multiple variables representing the state of the device, wherein, The device monitoring program is used to cause the computer to perform the following process: Based on the frequency distribution of a variable representing the state of the device, the range of the variable is divided into multiple first range bands; Based on the data of the plurality of variables corresponding to the plurality of second ranges of the variable determined based on the plurality of first ranges, a plurality of unit spaces are respectively constructed as the basis for the calculation of the Mahalanobis distance; as well as The value of the variable is selected from the plurality of first range bands to become the boundary between the plurality of second range bands.

Images