Anomaly diagnosis device, anomaly diagnosis method, and program.

Abstract

When an anomaly is detected, two sensors are selected from a number of sensors. The combinations include, when the SN ratios are sorted in descending order, sensors corresponding to SN ratios from a t

Description

Title of Invention: ANOMALY DIAGNOSIS DEVICE, ANOMALY DIAGNOSIS METHOD, AND PROGRAM Technical Field

[0001] The present disclosure relates to an anomaly diagnosis device, an anomaly diagnosis method, and a program. Background Art

[0002] As a method of remotely monitoring a plant such as a gas turbine, for example, a Mahalanobis-Taguchi method (hereinafter referred to as an MT method) has been used. In the MT method, while a plant as a monitoring target is operating in a normal state, a plurality of state quantities detected by various sensors installed in the monitoring target are acquired, and the acquired plurality of state quantities are set as a reference. Thereafter, state quantities detected by the various sensors are sequentially acquired over time, and a signal space is determined using the acquired state quantities. Every time a signal space is obtained, a difference between the signal space and a unit space determined using the state quantities defined as the reference is calculated as a Mahalanobis distance. For example, when the calculated Mahalanobis distance exceeds a predetermined threshold value, it is determined that an anomaly has occurred in the monitoring target.

[0003] In the MT method, a signal-to-noise (SN) ratio is further calculated as an index indicating the degree of influence on a Mahalanobis distance for each state quantity, that is, for each sensor. By using the SN ratio, it is possible to identify which sensor detects an anomaly in a state quantity. In other words, it is possible to identify, among portions of the monitoring target to be detected by the respective sensors, which portion has an anomaly (for example, refer to Patent Documents 1 and 2). Citation List Patent Literature

[0004] Patent Document 1: JP 2011-238148 A Patent Document 2: JP 2021-140540 A 05 06 25 5 Summary of Invention Technical Problem

[0005] A state quantity detected by the sensor having the largest SN ratio is one of factors that increase a Mahalanobis distance. However, various factors cause 10 anomalies in the monitoring target. For example, there may be a case where there is no cause for an anomaly at a part of a detection target of the sensor having the largest SN ratio, and a change in the state of a part causing the anomaly causes a change in the state quantity of the part of the detection target of the sensor having the largest SN ratio. 15

[0006] In such a case, an analyst who analyzes the anomaly needs to perform an analysis operation for identifying the cause of the anomaly while referring to a relationship between the state quantities from the plurality of sensors. However, since the number of sensors installed in the monitoring target is very 20 large, the number of combinations of sensors becomes enormous. Thus, there is a problem that it takes a lot of effort and time for the analyst to determine which combination of sensors should be referred to in the analysis operation for identifying the cause of the anomaly.

[0007] 25 The present invention has been made to solve the above-described problem, and an object thereof is to provide an anomaly diagnosis device, an abnormality diagnosis method, and a program that enable an analyst to determine a combination of sensors indicating a cause of an anomaly occurring in a monitoring target with little effort and time. 30 Solution to Problem

[0008] To solve the above-described problem, an anomaly diagnosis device, method and program according to claims 1, 10 and 11 are provided. The present 35 disclosure includes: an anomaly detection unit configured to detect an anomaly by applying a Mahalanobis-Taguchi method to state quantities detected by a plurality of sensors installed in a monitoring target and calculate SN ratios corresponding to the respective sensors; a sensor selection unit configured to select, when the anomaly detection unit detects the anomaly, combinations of 40 two sensors selected from the plurality of sensors, the combinations including, when the SN ratios are sorted in descending order, sensors corresponding to the SN ratios from a top rank to a first predetermined rank as one sensor and sensors which are different from the one sensor and satisfy a predetermined selection condition as the other sensor; and a display processing unit configured to select, as a first index, an index indicating a state quantity of one sensor included in each of the combinations of the two sensors selected by the sensor selection unit, select, as a second index, an index indicating a state quantity of the other sensor, and display, on a display screen, a scale of the first index, a scale of the second index, and positions of the respective state quantities determined by the scale of the first index and the scale of the second index per combination of the first index and the second index selected.

[0009] An anomaly diagnosis method according to the present disclosure includes: an anomaly detection step of detecting an anomaly by applying a Mahalanobis-Taguchi method to state quantities detected by a plurality of sensors installed in a monitoring target and calculating SN ratios corresponding to the respective sensors; a sensor selection step of selecting, when the anomaly is detected by the anomaly detection step, combinations of two sensors selected from the plurality of sensors, the combinations including, when the SN ratios are sorted in descending order, sensors corresponding to the SN ratios from a top rank to a first predetermined rank as one sensor and sensors which are different from the one sensor and satisfy a predetermined selection condition as the other sensor; and a display processing step of selecting, as a first index, an index indicating a state quantity of one sensor included in each of the combinations of the two sensors selected by the sensor selection step, selecting, as a second index, an index indicating a state quantity of the other sensor, and displaying, on a display screen, a scale of the first index, a scale of the second index, and positions of the respective state quantities determined by the scale of the first index and the scale of the second index per combination of the first index and the second index selected.

[0010] A program according to the present disclosure causes a computer to operate as: an anomaly detection means configured to detect an anomaly by applying a Mahalanobis-Taguchi method to state quantities detected by a plurality of sensors installed in a monitoring target and calculate SN ratios corresponding to the respective sensors; a sensor selection means configured to select, upon detection of the anomaly by the anomaly detection means combinations of two sensors selected from the plurality of sensors, the combinations including, when the SN ratios are sorted in descending order, sensors corresponding to the SN ratios from a top rank to a first predetermined rank as one sensor and sensors which are different from the one sensor and satisfy a predetermined selection condition as the other sensor; and a display processing means configured to select, as a first index, an index indicating a state quantity of one sensor included in each of the combinations of the two sensors selected by the sensor selection means, select, as a second index, an index indicating a state quantity of the other sensor as a second index, and display, on a display screen, a scale of the first index, a scale of the second index, and positions of the respective state quantities determined by the scale of the first index and the scale of the second index per combination of the first index and the second index selected. Advantageous Effects of Invention

[0011] According to the anomaly diagnosis device, the abnormality diagnosis method, and the program of the present disclosure, it is possible to enable an analyst to determine a combination of sensors indicating a cause of an anomaly occurring in a monitoring target with little effort and time. Brief Description of Drawings

[0012] FIG. lisa block diagram illustrating a configuration of an anomaly diagnosis device according to an embodiment of the present disclosure, a plant which is a monitoring target of the anomaly diagnosis device, and sensors and a display device connected to the anomaly diagnosis device. FIG. 2 is a diagram illustrating a data format of a table stored in a state quantity storage unit of the anomaly diagnosis device according to the embodiment of the present disclosure. FIG. 3 is a diagram illustrating a data format of a table stored in an SN ratio storage unit of the anomaly diagnosis device according to the embodiment of the present disclosure. FIG. 4 is a flowchart illustrating a processing flow by the anomaly diagnosis device according to the embodiment of the present disclosure. FIG. 5 is a diagram illustrating examples of graphs displayed on a display screen of the display device connected to the anomaly diagnosis device of the present disclosure. FIG. 6 is a schematic block diagram illustrating a configuration of a computer according to at least one embodiment. Description of Embodiments

[0013] Hereinafter, an anomaly diagnosis device, an anomaly diagnosis method, and a program according to an embodiment of the present disclosure will be described with reference to FIGS. 1 to 5. FIG. lisa block diagram illustrating a configuration of an anomaly diagnosis device 1 according to the embodiment of the present disclosure, a plant 30 which is a monitoring target of the anomaly diagnosis device 1, and sensors 20-1 to 20-N and a display device 2 connected to the anomaly diagnosis device 1. FIGS. 2 and 3 are diagrams respectively illustrating data formats of tables stored in a state quantity storage unit 12 and an SN ratio storage unit 14 of the anomaly diagnosis device 1 according to the embodiment of the present disclosure. FIG. 4 is a flowchart illustrating a processing flow by the anomaly diagnosis device 1 according to the embodiment of the present disclosure. FIG. 5 is a diagram illustrating examples of graphs displayed on a display screen of the display device 2 connected to the anomaly diagnosis device 1 of the present disclosure. Note that, in the drawings, the same reference signs are used for the same or corresponding configurations, and the descriptions thereof will be omitted as appropriate.

[0014] Configuration Example of Anomaly Diagnosis Device As illustrated in FIG. 1, the anomaly diagnosis device 1 is connected to the display device 2 and the sensors 20-1 to 20-N, and monitors a state of the plant 30 which is a monitoring target of the anomaly diagnosis device 1 with the sensors 20-1 to 20-N. The plant 30 is, for example, a gas turbine. The display device 2 is, for example, a liquid crystal display including a display screen.

[0015] The sensors 20-1 to 20-N are, for example, various sensors such as temperature sensors, pressure sensors, and voltage sensors, and are installed at portions of the plant 30 that need to be monitored. The value of “N” is an integer equal to or greater than two. Each of the sensors 20-1 to 20-N includes therein a time measuring means such as a clock, detects a state of a monitoring target portion in the plant 30 at which each of the sensors 20-1 to 20-N is installed at a constant time interval, and acquires a detected time (hereinafter, referred to as a detection time) from the time measuring means. Each of the sensors 20-1 to 20-N outputs state data including a state quantity indicating a quantity of a detected state, information indicating a detection time of the state quantity, and sensor identification information of each of the sensors 20-1 to 20-N. The sensor identification information is information assigned in advance to each of the sensors 20-1 to 20-N so that each of the sensors 20-1 to 20-N can be identified. Here, state quantities detected by the sensors 20-1 to 20-N are, for example, temperature values in the case of temperature sensors, pressure values in the case of pressure sensors, and voltage values in the case of voltage sensors. The information indicating the time obtained from the time measuring means provided inside each of the sensors 20-1 to 20-N is, for example, information represented in a format of year, month, day, hour, minute, and second.

[0016] The anomaly diagnosis device 1 includes a state quantity collection unit 11, a state quantity storage unit 12, an anomaly detection unit 13, an SN ratio storage unit 14, a sensor selection unit 15, and a display processing unit 16.

[0017] The state quantity collection unit 11 is connected to the sensors 20-1 to 20-N via, for example, an electric line or a communication line, and captures state data output by each of the sensors 20-1 to 20-N. The state quantity collection unit 11 corrects differences in detection time between the captured state data and associates the detection time with the state data such that N pieces of state data exist at one detection time. The state quantity collection unit 11 records N state quantities included in the N pieces of state data, sensor identification information corresponding to each of the N state quantities, and information indicating one detection time corresponding to the N state quantities as one record in the state quantity storage unit 12.

[0018] As illustrated in FIG. 2, the state quantity storage unit 12 stores a table in a data format having items of "Detection time", each of N pieces of "sensor identification information", and "Type". In the item of "Detection time", information indicating a detection time is recorded by the state quantity collection unit 11. In each of the N items of "sensor identification information", a state quantity corresponding to the sensor identification information of each item is recorded by the state quantity collection unit 11. In the item of "Type", any one of "Unit space", "Signal space", and "Anomaly" is recorded. When recording a record in the state quantity storage unit 12, the state quantity collection unit 11 records "Signal space" in the item of "Type".

[0019] A user of the anomaly diagnosis device 1, that is, an analyst who analyzes an anomaly, operates an operation unit (not illustrated) included in the anomaly diagnosis device 1 to select some records in a period in which the plant 30 operated normally from the records in the state quantity storage unit 12 and rewrites the items of "Type" of the selected records from "Signal space" to "Unit space". Note that it is desirable that, for example, the analyst periodically performs rewriting from "Signal space" to "Unit space" such that continuous records in a period of about two to three months in which the plant 30 operated normally most recently are rewritten from "Signal space" to "Unit space" and recorded in the state quantity storage unit 12.

[0020] The anomaly detection unit 13 determines a unit space using the state quantity of a record in which the item of "Type" is "Unit space" among the records recorded in the state quantity storage unit 12. The anomaly detection unit 13 continuously performs the processing of monitoring the processing of recording a new record in the state quantity storage unit 12 by the state quantity collection unit 11. When a new record is recorded in the state quantity storage unit 12, the anomaly detection unit 13 determines a signal space using the state quantity of the recorded new record.

[0021] The anomaly detection unit 13 calculates a Mahalanobis distance of the signal space by applying the MT method to the state quantity of the signal space and the state quantity of the unit space. When the calculated Mahalanobis distance exceeds a predetermined threshold value, the anomaly detection unit 13 determines that an anomaly has occurred in the plant 30. When it is determined that an anomaly has occurred in the plant 30, the anomaly detection unit 13 rewrites the item of "Type" in the record corresponding to the signal space at the time of determination of the anomaly from "Signal space" to "Anomaly".

[0022] The anomaly detection unit 13 calculates an SN ratio, which is an index indicating the degree of influence on the Mahalanobis distance for each of the sensors 20-1 to 20-N, based on the state quantity of the unit space and each of state quantities in a record, which is a determination target at the time of determination of the anomaly, of the state quantity storage unit 12. The anomaly detection unit 13 records, in the SN ratio storage unit 14, the calculated SN ratio for each of the sensors 20-1 to 20-N and information indicating the detection time in the record, which is the determination target at the time of determination of the anomaly, of the state quantity storage unit 12. Upon completion of recording in the SN ratio storage unit 14, the anomaly detection unit 13 outputs an anomaly occurrence notification signal including the information indicating the detection time recorded in the SN ratio storage unit 14.

[0023] As illustrated in FIG. 3, the SN ratio storage unit 14 stores a table in a data format having items of "Detection time" and each of N pieces of "sensor identification information". In the item of "Detection time", information indicating a detection time is recorded by the anomaly detection unit 13. In each of the N items of "sensor identification information", an SN ratio corresponding to the sensor identification information of each item is recorded by the anomaly detection unit 13.

[0024] The sensor selection unit 15 selects a combination of two sensors from among the plurality of sensors 20-1 to 20-N. The sensor selection unit 15 includes a sensor combination generation unit 15-1 and a two-dimensional Mahalanobis distance (MD) calculation unit 15-2. Upon capturing the anomaly occurrence notification signal output by the anomaly detection unit 13, the sensor combination generation unit 15-1 detects, from the SN ratio storage unit 14, a record corresponding to the information indicating the detection time included in the captured anomaly occurrence notification signal. The sensor combination generation unit 15-1 sorts N pieces of sensor identification information included in the detected record in descending order of SN ratio. The sensor combination generation unit 15-1 selects sensor identification information from the top rank to a predetermined rank (hereinafter referred to as a first predetermined rank).

[0025] The sensor combination generation unit 15-1 generates combined data in which the selected sensor identification information is included as one sensor identification information, and sensor identification information other than the one sensor identification information and corresponding to each of the sensors 20-1 to 20-N satisfying a predetermined selection condition is included as the other sensor identification information.

[0026] The two-dimensional MD calculation unit 15-2 reads the state quantities corresponding to the two pieces of sensor identification information included in the combined data generated by the sensor combination generation unit 15-1 from the state quantity storage unit 12, and determines a unit space and a signal space using the read state quantities. The two-dimensional MD calculation unit 15-2 calculates a two-dimensional Mahalanobis distance of the signal space based on the determined unit space and the determined signal space.

[0027] The two-dimensional MD calculation unit 15-2 calculates a two-dimensional Mahalanobis distance corresponding to each piece of combined data generated by the sensor combination generation unit 15-1, and then sorts the combined data in descending order of two-dimensional Mahalanobis distance. The two-dimensional MD calculation unit 15-2 selects combined data from the top rank to a predetermined rank (hereinafter referred to as a second predetermined rank).

[0028] The display processing unit 16 is connected to the display device 2 via, for example, an electric line. The display processing unit 16 reads the state quantity corresponding to each of the two pieces of sensor identification information included in the combined data selected by the two-dimensional MD calculation unit 15-2 from the state quantity storage unit 12. The display processing unit 16 selects an index indicating the state quantity of one sensor of the two pieces of sensor identification information included in the combined data selected by the two-dimensional MD calculation unit 15-2 as a first index and an index indicating the state quantity of the other sensor as a second index. The display processing unit 16 generates image data of a graph which is represented by a scale of the first index and a scale of the second index for each combination of the selected first index and the selected second index, and in which the position of each of the read state quantities is indicated by the scale of the first index and the scale of the second index. When generating the image data of the graph, the display processing unit 16 generates the image data of the graph such that the positions of the state quantities are indicated in a different form for each of the types of the read state quantities, that is, for the types of "Unit space", "Signal space ", and "Anomaly". The display processing unit 16 outputs the generated image data of the graph to the display device 2. The display device 2 displays the image of the graph on the display screen using the image data of the graph output by the display processing unit 16.

[0029] Processing by Anomaly Diagnosis Device FIG. 4 is a flowchart illustrating a flow of the processing performed by the sensor selection unit 15 and the display processing unit 16 after the anomaly detection unit 13 outputs an anomaly occurrence notification signal in the anomaly diagnosis device 1. In the following description, the predetermined selection condition is a condition of selecting all the sensors 20-1 to 20-N other than any of the sensors 20-1 to 20-N corresponding to the one sensor identification information, and the first predetermined rank is the top rank, and the second predetermined rank is the third rank.

[0030] The sensor combination generation unit 15-1 captures and acquires an anomaly occurrence notification signal output by the anomaly detection unit 13 (SI). The sensor combination generation unit 15-1 reads a record corresponding to the information indicating the detection time included in the acquired anomaly occurrence notification signal from the SN ratio storage unit 14. The sensor combination generation unit 15-1 sorts N pieces of sensor identification information included in the read record in descending order of SN ratio. The sensor combination generation unit 15-1 selects the sensor identification information of the top rank in the above-described order, that is, the sensor identification information having the largest SN ratio (S2).

[0031] The sensor combination generation unit 15-1 sets the selected sensor identification information having the largest SN ratio as sensor identification information serving as a primary element in generating a combination (hereinafter, referred to as primary sensor identification information). The sensor combination generation unit 15-1 generates N-l pieces of combined data each of which includes a primary sensor identification information as one sensor identification information and each of all pieces of sensor identification information other than the primary sensor identification information as the other sensor identification information in accordance with the predetermined selection condition. When generating the combined data, the sensor combination generation unit 15-1 assigns an identification label indicating the primary sensor identification information to the primary sensor identification information and generates the combined data.

[0032] The sensor combination generation unit 15-1 associates each of the generated N-l pieces of combined data with information indicating a detection time in the item "Detection time" of the record read from the SN ratio storage unit 14 when the N-l pieces of combined data are generated. The sensor combination generation unit 15-1 outputs the N-l pieces of combined data, each of which is associated with the information indicating the detection time, to the two-dimensional MD calculation unit 15-2 (S3).

[0033] The two-dimensional MD calculation unit 15-2 captures the N-l pieces of combined data output by the sensor combination generation unit 15-1, each of which is associated with the information indicating the detection time. The two-dimensional MD calculation unit 15-2 selects any one piece of combined data from the captured N-l pieces of combined data.

[0034] The two-dimensional MD calculation unit 15-2 reads all state quantities which correspond to the two pieces of sensor identification information included in the selected combined data and whose item "Type" is "Unit space" from the state quantity storage unit 12. The two-dimensional MD calculation unit 15-2 determines a unit space corresponding to the selected combined data by using the read state quantities whose item "Type" corresponds to "Unit space".

[0035] The two-dimensional MD calculation unit 15-2 detects one record corresponding to the information indicating the detection time associated with the selected combined data from the state quantity storage unit 12. The two-dimensional MD calculation unit 15-2 reads two state quantities respectively corresponding to the two pieces of sensor identification information included in the selected combined data from the detected one record. The two-dimensional MD calculation unit 15-2 determines a signal space corresponding to the selected combined data by using the read two state quantities. The two-dimensional MD calculation unit 15-2 calculates a two-dimensional Mahalanobis distance of the signal space based on the signal space and the determined unit space corresponding to the selected combined data. The two-dimensional MD calculation unit 15-2 further associates the selected combined data with the calculated two-dimensional Mahalanobis distance. The two-dimensional MD calculation unit 15-2 records the combined data associated with the information indicating the detection time and the two-dimensional Mahalanobis distance in an internal storage area (S4).

[0036] The two-dimensional MD calculation unit 15-2 repeatedly selects any one of the combined data that are not processed in S4 and performs the processing in S4 on the selected combined data until there is no combined data to be processed in S4 (loop from Lis to Lie). The processing in the flowchart of FIG. 4 is performed every time the anomaly detection unit 13 outputs an anomaly occurrence notification signal. To that end, during the processing in the first S4 in the processing of the loop from Lis to Lie, the two-dimensional MD calculation unit 15-2 initializes the internal storage area before recording the combined data associated with the information indicating the detection time and the two-dimensional Mahalanobis distance in the internal storage area. Accordingly, combined data at the time of a previous anomaly occurrence notification signal is cleared.

[0037] The two-dimensional MD calculation unit 15-2 reads the N-l pieces of combined data associated with the two-dimensional Mahalanobis distance and the information indicating the detection time from the internal storage area. The two-dimensional MD calculation unit 15-2 sorts the N-l pieces of combined data associated with the two-dimensional Mahalanobis distance and the information indicating the detection time in descending order of two-dimensional Mahalanobis distance. The two-dimensional MD calculation unit 15-2 selects three pieces of combined data from the top rank to the third rank. The two-dimensional MD calculation unit 15-2 further associates the selected three pieces of combined data, which are the combined data associated with the information indicating the detection time, with information indicating the rank, and outputs the three pieces of combined data to the display processing unit 16 (S5).

[0038] The display processing unit 16 captures the three pieces of combined data output by the two-dimensional MD calculation unit 15-2 which are the combined data associated with the information indicating the rank and the information indicating the detection time. For example, the display processing unit 16 performs processing of generating image data of a graph from each piece of the combined data in accordance with the ranks in the following procedure. That is, the display processing unit 16 first selects the combined data of the top rank, and reads the sensor identification information to which the identification label indicating the primary sensor identification information is assigned and the sensor identification information to which no identification label is assigned, both of the sensor identification information being included in the selected combined data. Here, it is assumed that the sensor corresponding to the sensor identification information to which the identification label is assigned is a sensor 20-A (where A is an integer equal to or greater than 1 and equal to or less than N) and the sensor corresponding to the sensor identification information to which no identification label is assigned is a sensor 20-P (where P is an integer equal to or greater than 1 and equal to or less than N and is an integer different from A).

[0039] The display processing unit 16 sets an index indicating the state quantity of the sensor 20-A as a first index and sets an index indicating the state quantity of the sensor 20-P as a second index. The display processing unit 16 reads the state quantities which correspond to the sensor identification information of the sensors 20-A and 20-P, respectively, and whose item "Type" is "Unit space" from the state quantity storage unit 12, and assigns a label indicating "Unit space" to the read state quantities.

[0040] The display processing unit 16 reads, from the state quantity storage unit 12, the state quantities which correspond to the sensor identification information of the sensors 20-A and 20-P, respectively, in which a date of the detection time in the item of "Detection time" matches the date indicated by the information indicating the detection time associated with the combined data of the top rank, and in which the item of "Type" is "Signal space" or "Anomaly". In other words, the display processing unit 16 reads, from the state quantity storage unit 12, the state quantities which correspond to the sensor identification information of the sensors 20-A and 20-P, respectively, which are detected on the same date as the date of the occurrence of the anomaly, and whose item "Type" is "Signal space" or "Anomaly".

[0041] When reading, from the state quantity storage unit 12, the state quantities which correspond to the sensor identification information of the sensors 20-A and 20-P, respectively, which are detected on the same date as the date of the occurrence of the anomaly, and whose item "Type" is "Unit space" or "Anomaly", the display processing unit 16 also reads the information of the item "Type" corresponding to the state quantities to be read, that is, the information of "Signal space" or "Anomaly". The display processing unit 16 assigns the read information of the item "Type" corresponding to the state quantities to the read state quantities.

[0042] Accordingly, the display processing unit 16 obtains state quantities which correspond to the combined data of the top rank, correspond to the sensor identification information of the sensors 20-A and 20-P, respectively, and each of which is assigned with the information of "Unit space", "Signal space", or "Anomaly".

[0043] The display processing unit 16 sets an index indicating the state quantity of the sensor 20-A as a first index and sets an index indicating the state quantity of the sensor 20-P as a second index. The display processing unit 16 generates image data of a graph of a scatter diagram in which the scale of the first index is set as the vertical axis, the scale of the second index is set as the horizontal axis, and positions of the respective state quantities determined by the scales of the first index and the second index and the types of the respective state quantities are indicated. In order to display the positions and the types of the state quantities, the display processing unit 16 generates the image data of the graph such that, for example, circular marks are displayed at the positions of the state quantities to which "Unit space" is assigned, rhombic marks are displayed at the positions of the state quantities to which "Signal space" is assigned, and star marks are displayed at the positions of the state quantities to which "Anomaly" is assigned. The display processing unit 16 outputs the generated image data of the graph to the display device 2. The display device 2 captures the image data of the graph output by the display processing unit 16 and displays an image of the graph on the display screen using the captured image data of the graph (S6). Accordingly, for example, the graph illustrated in FIG. 5(a) is displayed on the display screen of the display device 2.

[0044] Next, the display processing unit 16 selects the combined data of the second rank and performs the processing of S6. One sensor corresponding to the primary sensor identification information included in the combined data of the second rank is the sensor 20-A. Here, it is assumed that the other sensor included in the combined data of the second rank is a sensor 20-Q (where Q is an integer equal to or greater than 1 and equal to or less than N and is an integer different from A and P). In this case, the display processing unit 16 performs the processing of S6 on the combined data of the second rank to generate image data of a graph of a scatter diagram in which the first index corresponding to the sensor 20-A is set as the vertical axis and the second index corresponding to the sensor 20-Q is set as the horizontal axis. Accordingly, for example, the graph illustrated in FIG. 5(b) is displayed on the display screen of the display device 2.

[0045] Finally, the display processing unit 16 selects the combined data of the third rank and performs the processing of S6. One sensor corresponding to the primary sensor identification information included in the combined data of the third rank is the sensor 20-A. Here, it is assumed that the other sensor included in the combined data of the third rank is a sensor 20-R (where R is an integer equal to or greater than 1 and equal to or less than N and is an integer different from A, P, and Q). In this case, the display processing unit 16 performs the processing of S6 on the combined data of the third rank to generate image data of a graph of a scatter diagram in which the first index corresponding to the sensor 20-A is set as the vertical axis and the second index corresponding to the sensor 20-R is set as the horizontal axis. Accordingly, for example, the graph illustrated in FIG. 5(c) is displayed on the display screen of the display device 2 (loop from L2s to L2e). When the processing of the loop from L2s to L2e is completed, the processing of the flowchart illustrated in FIG. 4 is completed.

[0046] Operational Effect of Anomaly Diagnosis Device Upon completion of the processing of FIG. 4, three graphs illustrated in FIGS. 5(a), (b), and (c) are displayed on the display screen of the display device 2. The graph illustrated in FIG. 5(a) is a graph illustrating a relationship between the state quantity of the sensor 20-P having the largest two-dimensional Mahalanobis distance and the state quantity of the sensor 20-A in the relationship with the state quantity of the sensor 20-A. The graph illustrated in FIG. 5(b) is a graph illustrating a relationship between the state quantity of the sensor 20-Q having the second largest two-dimensional Mahalanobis distance and the state quantity of the sensor 20-A in the relationship with the state quantity of the sensor 20-A. The graph illustrated in FIG. 5(c) is a graph illustrating a relationship between the state quantity of the sensor 20-R having the third largest two-dimensional Mahalanobis distance and the state quantity of the sensor 20-A in the relationship with the state quantity of the sensor 20-A. Five star marks indicating the type "Anomaly" are displayed in the graphs of FIGS. 5(a), (b), and (c), indicating that these graphs are displayed on the display screen of the display device 2 when the anomaly detection unit 13 outputs the anomaly occurrence notification signal for the fifth time. By referring to these three graphs and analyzing how far the state quantities indicating the anomaly are away from the unit space, an analyst can identify a portion where the anomaly is occurring in the plant 30.

[0047] The number of combinations in which two sensors 20-1 and 20-N are selected from the N sensors 20-1 to 20-N is nC2, that is, N * (N-1) / 2. When the value of N increases, the number of combinations becomes enormous. On the other hand, the sensor selection unit 15 of the anomaly diagnosis device 1 narrows down the combinations to be displayed from the N * (N-l) / 2 combinations based on two indices of the magnitude of SN ratio and the magnitude of two-dimensional Mahalanobis distance. It can be said that the two indices of the magnitude of SN ratio and the magnitude of two-dimensional Mahalanobis distance are indices indicating a degree of relevance to an anomaly. Thus, the combinations narrowed down by the sensor selection unit 15 are combinations having a high degree of relevance to an anomaly, and the graphs displayed on the display screen of the display device 2 are graphs indicating the relationship of the combinations having the high degree of relevance to the anomaly. In other words, narrowing down the combinations based on the two indices of the magnitude of SN ratio and the magnitude of two-dimensional Mahalanobis distance can increase the possibility that a graph indicating the characteristics of the state quantities of the sensors 20-1 to 20-N that indicate a cause of an occurring anomaly is included in the graphs displayed on the display screen of the display device 2. Therefore, by using the anomaly diagnosis device 1, an analyst can identify a combination of sensors indicating a cause of an anomaly occurring in the plant 30 as a monitoring target with little effort and time.

[0048] Other Configuration Examples of Embodiment Other Configuration Examples of Predetermined Rank The first predetermined rank and the second predetermined rank on the premise of the processing in FIG. 4 are merely examples, and the first predetermined rank and the second predetermined rank may be any ranks. For example, when the top rank is set as the second predetermined rank, the two-dimensional MD calculation unit 15-2 outputs only the combined data having the largest two-dimensional Mahalanobis distance to the display processing unit 16, and thus, in the example of FIG. 4, the display processing unit 16 displays only the graph of FIG. 5(a) on the display screen of the display device 2.

[0049] When the rank indicated by the first predetermined rank is lowered, the number of pieces of combined data generated by the sensor combination generation unit 15-1 is increased. As a result, the processing load of the two dimensional MD calculation unit 15-2 is increased, but it is possible to extract a combination having a high degree of relevance to an occurring anomaly in a wider range. Therefore, the first predetermined rank is determined in view of an analysis range and the degree of the processing load of the two-dimensional MD calculation unit 15-2. When the first predetermined rank is set to the second rank or lower, the sensor combination generation unit 15-1 may select sensor identification information of any one or more ranks between the top rank and the first predetermined rank.

[0050] Since the second predetermined rank is related to the number of graphs to be displayed on the display screen of the display device 2 by the display processing unit 16, the display processing unit 16 displays more graphs on the display screen of the display device 2 as the rank indicated by the second predetermined rank is lowered. Accordingly, an analyst needs to perform an analysis by referring to more graphs, and thus the effort and time required for the analyst to perform the analysis are increased. However, since analysis targets of the analyst are increased, it is possible to increase the possibility of more accurately identifying a cause of an occurring anomaly. Thus, the second predetermined rank is determined in view of the effort and time required for an analyst to perform an analysis and a required accuracy. When the second predetermined rank is set to the second rank or lower, the two-dimensional MD calculation unit 15-2 may select combined data of any one or more ranks between the top rank and the second predetermined rank.

[0051] Other Configuration Example of Predetermined Selection Condition (First Example) The predetermined selection condition on the premise of the processing of FIG. 4 is merely an example, and a predetermined selection condition as described below may be applied. For example, as the predetermined selection condition, a condition of selecting any predetermined two or more sensors from the sensors 20-1 to 20-N other than the sensor 20-A corresponding to the primary sensor identification information may be applied. With this condition, the number of pieces of combined data generated by the sensor combination generation unit 15-1 can be reduced, and thus the processing load of the two-dimensional MD calculation unit 15-2 can be reduced.

[0052] Other Configuration Examples of Predetermined Selection Condition (Second Example) As the predetermined selection condition, a condition of selecting any of the sensors 20-1 to 20-N other than the sensor 20-A corresponding to the primary sensor identification information may be applied, any of the sensors indicating a high degree of correlation between the state quantity of the sensor 20-A and the state quantities of the sensors 20-1 to 20-N other than the sensor 20-A with respect to a relationship with a predetermined threshold value. A specific process to be performed when this condition is applied will be described. For example, in the processing in S3 of FIG. 4, after determining the primary sensor identification information, the sensor combination generation unit 15-1 reads, from the state quantity storage unit 12, a state quantity of the primary sensor identification information in which the item "Type" is "Unit space". The sensor combination generation unit 15-1 selects any one piece of sensor identification information other than the primary sensor identification information, and reads, from the state quantity storage unit 12, a state quantity of the selected sensor identification information in which the item "Type" is "Unit space". The sensor combination generation unit 15-1 calculates a correlation coefficient based on the state quantity of the unit space of the primary sensor identification information and the state quantity of the unit space of the selected sensor identification information.

[0053] The sensor combination generation unit 15-1 calculates a correlation coefficient corresponding to each piece of sensor identification information other than the primary sensor identification information in the above-described procedure. The sensor combination generation unit 15-1 selects sensor identification information whose absolute value of the correlation coefficient exceeds a predetermined threshold value. The sensor combination generation unit 15-1 generates combined data including the primary sensor identification information as one sensor identification information and each piece of the sensor identification information whose absolute value of the correlation coefficient exceeds the predetermined threshold value as the other sensor identification information. The sensor combination generation unit 15-1 associates each piece of the generated combined data with information indicating a detection time in the item "Detection time" of a record in the SN ratio storage unit 14 which has been referred to in generating the combined data. The sensor combination generation unit 15-1 outputs the combined data associated with the information indicating the detection time to the two-dimensional MD calculation unit 15-2. Then, the processing of the loop from Lis to Lie is performed.

[0054] Accordingly, the sensor combination generation unit 15-1 can generate combined data in which the primary sensor identification information selected based on the magnitude of SN ratio and having a high degree of relevance to an occurring anomaly is set as one sensor identification information and sensor identification information of the sensors 20-1 to 20-N that detect state quantities indicating a high degree of correlation with the state quantity corresponding to the primary sensor identification information is set as the other sensor identification information.

[0055] In this case, since the sensor combination generation unit 15-1 narrows down the number of pieces of sensor identification information to be selected as the other sensor identification information as compared with a case where all pieces of sensor identification information other than the primary sensor identification information are selected as the other sensor identification information, it is possible to reduce the number of pieces of combined data to be generated. Thus, it is possible to reduce the processing load of the two-dimensional MD calculation unit 15-2. Further, the sensor combination generation unit 15-1 does not randomly narrow down the number of pieces of sensor identification information to be selected as the other sensor identification information, but narrows down to the sensor identification information of the sensors 20-1 to 20-N that detect state quantities having a high degree of correlation with the state quantity of the sensor 20-1. Accordingly, it is possible to increase the possibility that a graph indicating the characteristics of the state quantities of the sensors 20-1 to 20-N that indicate a cause of an occurring anomaly is included in the graphs displayed on the display screen of the display device 2 by the display processing unit 16 in the processing in S6.

[0056] The sensor combination generation unit 15-1 may calculate a determination coefficient instead of calculating a correlation coefficient. A determination coefficient is represented by a numerical value from 0 to 1, and thus it is not necessary to compare the absolute value thereof with a threshold value unlike a correlation coefficient which is represented by a numerical value from -1 to 1, and it is sufficient to compare the determination coefficient with the threshold value.

[0057] Other Configuration Examples of Predetermined Selection Condition (Third Example) As the predetermined selection condition, a condition of selecting any of the sensors 20-1 to 20-N other than the sensor 20-A corresponding to the primary sensor identification information may be applied, any of the sensors being predetermined with respect to the sensor 20-A corresponding to the primary sensor identification information. For example, for each of the sensors 20-1 to 20-N, other sensors 20-1 to 20-N estimated to have a high degree of relevance such as other sensors 20-1 to 20-N having a large two-dimensional Mahalanobis distance and other sensors 20-1 to 20-N useful for identifying a cause of an anomaly can be determined in advance based on experience or the like regarding anomalies occurred in the past. List data indicating the correlation determined in this manner between each of the sensors 20-1 to 20-N and the other sensors 20-1 to 20-N corresponding to each of the sensors 20-1 to 20-N by sensor identification information is generated in advance. The generated list data is stored in advance in an internal storage area of the sensor combination generation unit 15-1.

[0058] In the processing in S3 of FIG. 4, after determining the primary sensor identification information, the sensor combination generation unit 15-1 refers to the list data stored in the internal storage area and reads sensor identification information associated with the primary sensor identification information in the list data. The sensor combination generation unit 15-1 generates combined data including the primary sensor identification information as one sensor identification information and each piece of sensor identification information read from the list data as the other sensor identification information. The sensor combination generation unit 15-1 associates each piece of the generated combined data with information indicating a detection time in the item "Detection time" of a record in the SN ratio storage unit 14 which has been referred to in generating the combined data. The sensor combination generation unit 15-1 outputs the combined data associated with the information indicating the detection time to the two-dimensional MD calculation unit 15-2. Then, the processing of the loop from Lis to Lie is performed.

[0059] Accordingly, the sensor combination generation unit 15-1 can generate combined data in which the primary sensor identification information selected based on the magnitude of SN ratio and having a high degree of relevance to an occurring anomaly is set as one sensor identification information and sensor identification information of the sensors 20-1 to 20-N estimated to have a high degree of relevance to the sensor 20-A corresponding to the primary sensor identification information is set as the other sensor identification information.

[0060] In this case, since the sensor combination generation unit 15-1 narrows down the number of pieces of sensor identification information to be selected as the other sensor identification information as compared with a case where all pieces of sensor identification information other than the primary sensor identification information are selected as the other sensor identification information, it is possible to reduce the number of pieces of combined data to be generated. Thus, it is possible to reduce the processing load of the two-dimensional MD calculation unit 15-2. Further, the sensor combination generation unit 15-1 does not randomly narrow down the number of pieces of sensor identification information to be selected as the other sensor identification information, but narrows down to the sensor identification information of the sensors 20-1 to 20-N estimated to have a high degree of relevance to the sensor 20-A of the primary sensor identification information. Accordingly, it is possible to increase the possibility that a graph indicating the characteristics of the state quantities of the sensors 20-1 to 20-N that indicate a cause of an occurring anomaly is included in the graphs displayed on the display screen of the display device 2 by the display processing unit 16 in the processing in S6.

[0061] When either one of the predetermined selection condition of the second example of the other configuration examples of the predetermined selection condition and the predetermined selection condition of the third example of the other configuration examples of the predetermined selection condition is applied as the predetermined selection condition, the sensor combination generation unit 15-1 may generate combined data using the sensor identification information corresponding to any of the sensor 20-1 to 20-N arbitrarily selected from the sensors 20-1 to 20-N satisfying the predetermined selection condition instead of using the sensor identification information corresponding to all of the sensors 20-1 to 20-N satisfying the predetermined selection condition as described above.

[0062] Alternatively, the sensor combination generation unit 15-1 may apply both of the predetermined selection condition of the second example of the other configuration examples of the predetermined selection condition and the predetermined selection condition of the third example of the other configuration examples of the predetermined selection condition as the predetermined selection condition, and may generate combined data using the sensor identification information corresponding to all of the sensors 20-1 to 20-N satisfying the predetermined selection condition, or may generate combined data using the sensor identification information corresponding to any of the sensor 20-1 to 20-N arbitrarily selected from the sensors 20-1 to 20-N satisfying the predetermined selection condition.

[0063] Type of Graph In the embodiment described above, the display processing unit 16 indicates the types "Unit space", "Signal space", and "Anomaly" of state quantities by marks having different shapes. Alternatively, the display processing unit 16 may display a graph in any form as long as differences between the types "Unit space", "Signal space", and "Anomaly" of state quantities can be recognized, for example, by indicating the types "Unit space", "Signal space", and "Anomaly" of state quantities by marks with different colors. In that case, the display processing unit 16 does not necessarily use a form in which differences between the three types "Unit space", "Signal space", and "Anomaly" can be recognized. For example, the display processing unit 16 may regard state quantities of "Anomaly" as part of state quantities of "Signal space" and display state quantities in a form in which a difference between the state quantities of "Unit space" and state quantities of "Signal space" can be recognized, or may display only positions of state quantities without considering the types "Unit space", "Signal space", and "Anomaly".

[0064] In the embodiment described above, as illustrated in FIG. 5, the display processing unit 16 displays a graph of a scatter diagram in which an index indicating the state quantity of the sensor 20-A corresponding to the primary sensor identification information to which an identification label is assigned is set as a first index, the first index is represented by the vertical axis, and a second index is represented by the horizontal axis. Alternatively, a graph of a scatter diagram in which the first index is represented by the horizontal axis and the second index is represented by the vertical axis may be used. The sensor combination generation unit 15-1 may generate combined data without assigning an identification label, and the display processing unit 16 may set an index indicating the state quantity of any of the sensors 20-1 to 20-N corresponding to one sensor identification information randomly selected from two pieces of sensor identification information included in the combined data as the first index, and set an index indicating the state quantity of any of the sensors 20-1 to 20-N corresponding to the other sensor identification information as the second index. The display processing unit 16 may generate image data of a graph other than a scatter diagram as long as the graph indicates the scales of the first index and the second index, the positions of the state quantities determined by the scales of the first index and the second index, and the types of the state quantities.

[0065] Display of Graph In the above-described embodiment, the processing of FIG. 4 is performed every time the anomaly detection unit 13 outputs an anomaly occurrence notification signal. In this case, when the anomaly detection unit 13 outputs a plurality of anomaly occurrence notification signals in a short period of time, three graphs are additionally displayed for each abnormality occurrence notification signal on the display screen of the display device 2. In order to prevent many graphs from being displayed on the display screen of the display device 2 as in the above case, for example, the display processing unit 16 may perform processing of selectively displaying graphs on the display screen of the display device 2 as described below.

[0066] The display processing unit 16 captures combined data which is output by the two-dimensional MD calculation unit 15-2 and is associated with information indicating a rank and information indicating a detection time, and records the captured combined data in the internal storage area. An analyst performs an operation of designating a detection time on the display processing unit 16 via an operation unit (not illustrated) included in the anomaly diagnosis device 1. The display processing unit 16 reads three pieces of combined data corresponding to the detection time received from the operation unit from the internal storage area, and performs the processing of the loop from L2s to L2e. Accordingly, only the images of three graphs corresponding to the detection time designated by the analyst can be displayed on the display screen of the display device 2.

[0067] The analyst may designate a detection time and a rank to the display processing unit 16 via the operation unit (not illustrated). In that case, the display processing unit 16 reads one piece of combined data corresponding to the detection time and the rank received from the operation unit from the internal storage area, and performs the processing in S6. Accordingly, only the image of one graph corresponding to the detection time and the rank designated by the analyst can be displayed on the display screen of the display device 2.

[0068] Two-Dimensional MD Calculation Unit In the anomaly diagnosis device 1 of the above-described embodiment, the sensor selection unit 15 may include only the sensor combination generation unit 15-1 and may not include the two-dimensional MD calculation unit 15-2. In that case, the display processing unit 16 displays a graph on the display screen of the display device 2 based on combined data generated by the sensor combination generation unit 15-1. Thus, although the anomaly diagnosis device 1 cannot exert a narrowing down effect using two-dimensional Mahalanobis distances, it is possible to eliminate the processing load of the two-dimensional MD calculation unit 15-2. In that case, as the predetermined selection condition applied in the sensor combination generation unit 15-1, any of the above-described predetermined selection conditions may be applied.

[0069] Other Embodiments The embodiment according to the present disclosure has been described in detail with reference to the drawings. However, the specific configuration of the present disclosure is not limited to this embodiment, and also includes a design change or the like without departing from the gist of the present disclosure.

[0070] Computer Configuration FIG. 6 is a schematic block diagram illustrating a configuration of a computer according to at least one embodiment. A computer 90 includes a processor 91, a main memory 92, a storage 93, and an interface 94. The sensors 20-1 to 20-N and the display device 2 are connected to the interface 94. The above-described anomaly diagnosis device 1 is implemented in the computer 90. Operations of the state quantity collection unit 11, the anomaly detection unit 13, the sensor selection unit 15, and the display processing unit 16 of the anomaly diagnosis device 1 are stored in the form of program in the storage 93. The processor 91 reads the program from the storage 93, loads the program in the main memory 92, and performs the above-described processing in accordance with the program. Further, the processor 91 secures, in the main memory 92, storage areas corresponding to the internal storage areas of the sensor combination generation unit 15-1 and the two-dimensional MD calculation unit 15-2, the state quantity storage unit 12, and the SN ratio storage unit 14 described above. The storage areas of the state quantity storage unit 12 and the SN ratio storage unit 14 may be secured in the storage 93.

[0071] The program may be a program for achieving part of the functions achieved by the computer 90. For example, the program may be a program that achieves a function in combination with another program already stored in the storage 93, or in combination with another program installed in another device. Note that, in other embodiments, the computer may include a custom large scale integrated circuit (LSI) such as a programmable logic device (PLD), in addition to or in place of the configuration described above. Examples of the PLD include a programmable array logic (PAL), a generic array logic (GAL), a complex programmable logic device (CPLD), and a field programmable gate array (FPGA). In this case, some or all of the functions implemented by the processor may be implemented by the integrated circuit.

[0072] Other examples of the storage 93 include a hard disk drive (HDD), a solid state drive (SSD), a magnetic disk, a magneto-optical disk, a compact disc read only memory (CD-ROM), a digital versatile disc read only memory (DVD-ROM), and a semiconductor memory. The storage 93 may be an internal medium directly connected to a bus of the computer 90, or may be an external medium connected to the computer 90 via the interface 94 or a communication line. In addition, in a case where the program is distributed to the computer 90 through a communication line, the computer 90 that has received the distribution may load the program in the main memory 92 to perform the above-described processing. In at least one embodiment, the storage 93 is a non-temporary tangible storage medium.

[0073] Supplementary Notes The anomaly diagnosis device 1 according to the above-described embodiments is understood as follows, for example.

[0074] (1) An anomaly diagnosis device 1 according to a first aspect includes: an anomaly detection unit 13 configured to detect an anomaly by applying a Mahalanobis-Taguchi method to state quantities detected by a plurality of sensors 20-1 to 20-N installed in a monitoring target and calculate SN ratios corresponding to the respective sensors 20-1 to 20-N; a sensor selection unit 15 configured to select, when the anomaly detection unit detects the anomaly 13, combinations of two sensors selected from the plurality of sensors 20-1 to 20-N, the combinations including, when the SN ratios are sorted in descending order, the sensors 20-1 to 20-N corresponding to the SN ratios from a top rank to a first predetermined rank as one sensor and the sensors 20-1 to 20-N which are different from the one sensor and satisfy a predetermined selection condition as the other sensor; and a display processing unit 16 configured to select, as a first index, an index indicating a state quantity of the one sensor 20-1 to 20-N included in each of the combinations of two sensors selected by the sensor selection unit 15, select, as a second index, an index indicating a state quantity of the other sensor 20-1 to 20-N, and display, on a display screen, a scale of the first index, a scale of the second index, and positions of the respective state quantities determined by the scale of the first index and the scale of the second index per combination of the first index and the second index selected. According to the present aspect and the following aspects, it is possible to enable an analyst to determine a combination of sensors indicating a cause of an anomaly occurring in a monitoring target with little effort and time.

[0075] (2) An anomaly diagnosis device 1 according to a second aspect is the anomaly diagnosis device 1 of (1), wherein the sensor selection unit 15 calculates two-dimensional Mahalanobis distances from state quantities of two sensors included in the combinations for each of the combinations of two sensors, sorts the two-dimensional Mahalanobis distances in descending order, and newly selects the combinations of two sensors from among the combinations corresponding to the two-dimensional Mahalanobis distances from a top rank to a second predetermined rank. According to the present aspect, it is possible to narrow down the combinations of two sensors based on the magnitude of two-dimensional Mahalanobis distance.

[0076] (3) An anomaly diagnosis device 1 according to a third aspect is the anomaly diagnosis device 1 according to (1) or (2), wherein the predetermined selection condition is a condition of selecting all the sensors 20-1 to 20-N other than the one sensor. According to the present aspect, it is possible to generate the combinations of two sensors by selecting all sensors that can be combined with the one sensor selected based on the SN ratio.

[0077] (4) An anomaly diagnosis device 1 according to a fourth aspect is the anomaly diagnosis device 1 according to (1) or (2), wherein the predetermined selection condition is a condition of selecting any predetermined two or more sensors 20-1 to 20-N other than the one sensor. According to the present aspect, the other sensor of the combinations of the two sensors can be narrowed down to a predetermined plurality of sensors 20-1 to 20-N.

[0078] (5) An anomaly diagnosis device 1 according to a fifth aspect is the anomaly diagnosis device 1 according to (1) or (2), wherein the predetermined selection condition is a condition of selecting the sensors 20-1 to 20-N other than the one sensor indicating a high degree of correlation between the state quantity of the one sensor and state quantities of the sensors other than the one sensor with respect to a relationship with a predetermined threshold value. According to the present aspect, it is possible to narrow down the combinations of two sensors based on the degree of correlation.

[0079] (6) An anomaly diagnosis device 1 according to a sixth aspect is the anomaly diagnosis device 1 according to any one of (1), (2), and (5), wherein the predetermined selection condition is a condition of selecting the sensors 20-1 to 20-N other than the one sensor predetermined with respect to the one sensor. According to the present aspect, the other sensor of the combinations of two sensors can be narrowed down to the predetermined sensors 20-1 to 20-N.

[0080] (7) An anomaly diagnosis device 1 according to a seventh aspect is the anomaly diagnosis device 1 according to any one of (1) to (6), wherein the display processing unit 16 displays, when displaying positions of the respective state quantities determined by the scale of the first index and the scale of the second index, the positions of the state quantities in a different form per type of the state quantities. According to the present aspect, since the positions of the state quantities are displayed on the display screen in a different form for each type of the state quantities, for example, it is possible to distinguish whether each of the state quantities is a state quantity corresponding to a unit space, a state quantity corresponding to a signal space, or a state quantity at the time of occurrence of an anomaly on the display screen.

[0081] (8) An anomaly diagnosis device 1 according to an eighth aspect is the anomaly diagnosis device 1 according to any one of (1) to (7), wherein the first predetermined rank is a top rank. According to the present aspect, the one sensor selected based on the SN ratio can be a sensor having the largest SN ratio.

[0082] (9) An anomaly diagnosis device 1 according to a ninth aspect is the anomaly diagnosis device 1 according to any one of (2) to (8), wherein the second predetermined rank is a top rank. According to the present aspect, the combination of two sensors displayed on the display screen can be set to the combination having the largest two-dimensional Mahalanobis distance. Reference Signs List

[0083] 1 Anomaly diagnosis device 2 Display device 11 State quantity collection unit 12 State quantity storage unit 13 Anomaly detection unit 14 SN ratio storage unit 15 Sensor selection unit 15-1 Sensor combination generation unit 15-2 Two-dimensional MD calculation unit 16 Display processing unit 20-1 to 20-N Sensor 30 Plant 05 06 25

Claims

1. An anomaly diagnosis device, comprising:5 an anomaly detection unit configured to detect an anomaly by applying aMahalanobis-Taguchi method to state quantities detected by a plurality of sensors installed in a monitoring target and calculate SN ratios corresponding to the respective sensors;a sensor selection unit configured to select, when the anomaly detection 10 unit detects the anomaly, combinations of two sensors selected from the plurality of sensors, the combinations including, when the SN ratios are sorted in descending order, sensors corresponding to the SN ratios from a top rank to a first predetermined rank as one sensor and sensors which are different from the one sensor, detect state quantities of a possible cause of the anomaly, and 15 satisfy a predetermined selection condition as the other sensor; anda display processing unit configured to select, as a first index, an index indicating a state quantity of one sensor included in each of the combinations of the two sensors selected by the sensor selection unit, select, as a second index, an index indicating a state quantity of the other sensor, and display, on a 20 display screen, a scale of the first index, a scale of the second index, and positions of the respective state quantities determined by the scale of the first index and the scale of the second index per combination of the first index and the second index selected.25

2. The anomaly diagnosis device according to claim 1, wherein the sensor selection unit calculates two-dimensional Mahalanobis distances from state quantities of two sensors included in the combinations for each of the combinations of two sensors, sorts the two-dimensional Mahalanobis distances 30 in descending order, and newly selects the combinations of two sensors from among the combinations corresponding to the two-dimensional Mahalanobis distances from a top rank to a second predetermined rank.

3. 35 The anomaly diagnosis device according to claim 1 or 2, wherein thepredetermined selection condition is a condition of selecting all the sensors other than the one sensor.05 06 25

4. The anomaly diagnosis device according to claim 1 or 2, wherein the predetermined selection condition is a condition of selecting any predetermined two or more sensors other than the one sensor.5

5. The anomaly diagnosis device according to claim 1 or 2, wherein the predetermined selection condition is a condition of selecting sensors other than the one sensor indicating a high degree of correlation between the state quantity 10 of the one sensor and state quantities of the sensors other than the one sensor with respect to a relationship with a predetermined threshold value.

6. The anomaly diagnosis device according to any one of claims 1, 2, and 15 5, wherein the predetermined selection condition is a condition of selectingsensors which are different from the one sensor and predetermined with respect to the one sensor.

7. 20 The anomaly diagnosis device according to any one of claims 1 to 6,wherein the display processing unit displays, when displaying the positions of the respective state quantities determined by the scale of the first index and the scale of the second index, the positions of the state quantities in a different form per type of the state quantities25 wherein the type of the state quantities includes a state quantitycorresponding to a unit space, a state quantity corresponding to a signal space, or a state quantity at the time of occurrence of an anomaly.

8. 30 The anomaly diagnosis device according to any one of claims 1 to 7,wherein the first predetermined rank is a top rank when the SN ratios are sorted in descending order.

9. 35 The anomaly diagnosis device according to any one of claims 2 to 8,wherein the second predetermined rank is a top rank when the two-dimensional Mahalanobis distances are sorted in descending order.05 06 25

10. An anomaly diagnosis method comprising:an anomaly detection step of detecting an anomaly by applying a Mahalanobis-Taguchi method to state quantities detected by a plurality of5 sensors installed in a monitoring target and calculating SN ratios corresponding to the respective sensors;a sensor selection step of selecting, when the anomaly is detected by the anomaly detection step, combinations of two sensors selected from the plurality of sensors, the combinations including, when the SN ratios are sorted in10 descending order, sensors corresponding to the SN ratios from a top rank to a first predetermined rank as one sensor and sensors which are different from the one sensor, detect state quantities of a possible cause of the anomaly, and satisfy a predetermined selection condition as the other sensor; anda display processing step of selecting an index indicating, as a first15 index, a state quantity of the one sensor included in each of the combinations of two sensors selected in the sensor selection step, selecting, as a second index, an index indicating a state quantity of the other sensor, and displaying, on a display screen, a scale of the first index, a scale of the second index, and positions of the respective state quantities determined by the scale of the first20 index and the scale of the second index per combination of the first index and the second index selected.

11. A program causing a computer to operate as:25 an anomaly detection means configured to detect an anomaly byapplying a Mahalanobis-Taguchi method to state quantities detected by a plurality of sensors installed in a monitoring target and calculate SN ratios corresponding to the respective sensors;a sensor selection means configured to select, upon detection of the30 anomaly by the anomaly detection means, combinations of two sensors selected from the plurality of sensors, the combinations including, when the SN ratios are sorted in descending order, sensors corresponding to the SN ratios from a top rank to a first predetermined rank as one sensor and sensors which are different from the one sensor, detect state quantities of a possible cause of the35 anomaly, and satisfy a predetermined selection condition as the other sensor; anda display processing means configured to select, as a first index, an index indicating a state quantity of the one sensor included in each of thecombinations of two sensors selected by the sensor selection means, select, as a second index, an index indicating a state quantity of the other sensor, and display, on a display screen, a scale of the first index, a scale of the second index, and positions of the respective state quantities determined by the scale5 of the first index and the scale of the second index per combination of the first index and the second index selected.06 25

Images