ANOMAL FACTOR CALCULATION DEVICE, LEARNING DEVICE, PRECISE DIAGNOSTIC SYSTEM AND ANOMAL FACTOR CALCULATION METHOD

Abstract

Included are: a sensor data acquisition unit (10) for acquiring a plurality of parts of time-series sensor data collected from a plurality of sensors (300) provided in a plurality of plant components forming a target plant; an anomaly detection unit (30) for detecting an anomaly in a plurality of anomaly detection sensors based on a plurality of sensor data parts; an anomaly detection sequence calculation unit (40) for calculating an anomaly detection sequence in which the occurrence of the anomaly is recorded for a plurality of anomaly detection sensors; and an anomaly propagation path tracking unit (50) for calculating an anomaly propagation sequence in which the anomaly has propagated based on anomaly detection sensor information relating to a plurality of anomaly detection sensors and a calculated structure.which indicates a dependency relationship between the plant components, and an anomaly factor calculation unit (60) for calculating an anomaly factor based on the anomaly detection sequence and the anomaly propagation sequence.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to an anomaly factor calculation device, a learning device, a precise diagnostic system and an anomaly factor calculation method. BACKGROUND TO THE STATE OF THE ART

[0002] In a facility or plant, such as a factory or plant, several elements (hereinafter referred to as "plant components") that constitute the plant, such as various devices, interact with one another, and sensor data (variables) acquired by sensors integrated into and attached to the plant components are also of some importance. Therefore, if an anomaly occurs in a particular plant component among the multiple plant components that make up the plant, and the anomaly is detected, its influence spreads, and the anomaly is recorded in a multitude of sensor data. In such a case, it is not easy to specify (precisely diagnose) the plant component that is the source of the anomaly.

[0003] Therefore, a technique is usually known which, if a majority of plant components in a plant are in an abnormal operating state, can estimate or calculate a plant component that caused the abnormal operating state using sensor data collected by sensors provided in the majority of plant components in the plant.

[0004] For example, patent literature 1 discloses an anomaly diagnostic system that estimates or determines a part causing a change in the state of a plant based on a change in state that is based on a change in a relationship between a plurality of operating data relating to a target part, which are recorded by a plurality of sensing units set for each part (device) of the plant, and information of the relationship between the sensing units in which a propagation relationship of an influence between parts of the plant corresponding to the sensing units is stored. REFERENCE LISTPATENT LITERATURE

[0005] Patent literature 1: WO 2017 / 159016 A1 SUMMARY OF THE INVENTIONTECHNICAL PROBLEM

[0006] In a related prior art, as disclosed in patent literature 1, it is necessary for an operator or similar to detect a propagation relationship of an influence between a plurality of plant components in advance and to be able to prepare information that corresponds to information about the relationship between detection units, wherein the propagation relationship is defined on the basis of the detected propagation relationship.

[0007] On the other hand, in a complex or large-scale system where feedback control is implemented, it is difficult for an operator or similar to grasp the propagation relationship of an influence between the devices that make up the system or facility.

[0008] Therefore, the problem in the state of the art is that a propagation relationship of an influence between a plurality of plant components cannot be recorded, or even if the propagation relationship can be recorded, it is possible that the estimation or calculation of a factor of the anomaly that has occurred in the plant is not possible due to low accuracy.

[0009] The present disclosure was made to solve the above-mentioned problems, and one objective of it is to provide an anomaly factor calculation device that is capable of estimating or calculating a factor of an anomaly that has occurred in a plant, regardless of the complexity or size of the plant. SOLUTION TO THE PROBLEM

[0010] An anomaly factor calculation device according to the present disclosure comprises a sensor data acquisition unit for acquiring a plurality of parts of time-series sensor data acquired by a plurality of sensors provided in a plurality of plant components forming a target plant, an anomaly detection unit for acquiring a plurality of anomaly detection sensors in which an anomaly has occurred among a plurality of sensors, based on a plurality of sensor data acquired by the sensor data acquisition unit, and an anomaly detection sequence calculation unit for calculating orEstimating an anomaly detection sequence in which the occurrence of the anomaly is recorded for a plurality of anomaly detection sensors, based on a detection time at which the anomaly detection unit has recorded a plurality of the anomaly detection sensors; an anomaly propagation path tracking unit for calculating or estimating an anomaly propagation sequence in which the anomaly has propagated, based on anomaly detection sensor information regarding a plurality of the anomaly detection sensors recorded by the anomaly detection unit; a calculated structure that indicates a dependency relationship between the plant components; and an anomaly factor calculation unit for calculating or estimatingEstimating an anomaly factor based on the anomaly detection sequence calculated by the anomaly detection sequence calculation unit and the anomaly propagation sequence calculated by the anomaly propagation path tracking unit. ADVANTAGEOUS EFFECTS OF THE INVENTION

[0011] According to the present disclosure, the anomaly factor calculation device with the above configuration can calculate or estimate the factor of the anomaly that has occurred in the plant, regardless of the complexity or size of the plant. BRIEF DESCRIPTION OF THE DRAWINGS Fig. Figure 1 is a block diagram showing a configuration example of a precise diagnostic system with an anomaly factor calculation device according to a first embodiment. Fig. Figure 2 is a block diagram showing a configuration example of the anomaly factor calculation device according to the first embodiment. Fig. Figure 3 is a representation describing a configuration of sensor data in the first embodiment. Fig. Figure 4 is a representation describing a specific example of the calculation of the anomaly propagation sequence performed by an anomaly propagation path tracking unit in the first embodiment. Fig. Figure 5 is a further illustration to describe a specific example of the anomaly propagation sequence calculation processing performed by the anomaly propagation path tracking unit in the first embodiment. Fig. Figure 6 is a further illustration to describe a specific example of the anomaly propagation sequence calculation processing performed by the anomaly propagation path tracking unit in the first embodiment. Fig. Figure 7 is a representation to describe a specific example of an anomaly factor calculation processing performed by an anomaly factor calculation unit in the first embodiment. Fig. Figure 8 is a representation to describe a screen example of an anomaly factor calculation result screen, which is displayed on a display device by an anomaly factor calculation result output unit in the first embodiment. Fig. Figure 9 is a representation to describe another screen example of the anomaly factor calculation result screen, which is displayed on the display device by the anomaly factor calculation result output unit in the first embodiment. Fig. Figure 10A is a representation that provides an example of the content of an anomaly detection sequence calculation result. Fig. Figure 10B is a representation that provides an example of the content of an anomaly propagation sequence calculation result, and Fig. 10C is a representation that provides an example of the content of an anomaly factor order calculation result. Fig. Figure 11 is a block diagram showing a configuration example of a learning device according to the first embodiment. Fig. Figure 12 is a representation illustrating a concept of an example of learning processing in which a reference structure learning unit learns a reference structure or reference structure in the first embodiment. Fig. Figure 13 is a flowchart describing the operation of the anomaly factor calculation device according to the first embodiment. Fig. Figure 14 is a flowchart describing the operation of the learning device according to the first embodiment. Fig. 15 is a flowchart describing details of the processing in step ST23 in Fig. 14. Fig. 16A and Fig. Figure 16B shows block diagrams illustrating an example of a hardware configuration of the anomaly factor calculation device 100 according to the first embodiment. Fig. Figure 17 is a block diagram illustrating a configuration example of a precise diagnostic system in which the anomaly factor calculation device and the learning device include a sensor data acquisition unit and a data storage unit, which are common in the first embodiment. Fig. Figure 18 is a block diagram illustrating a configuration example of the precise diagnostic system, in which, in the first embodiment, the learning device learns the reference structure for each operating state of a target plant or facility, and the anomaly factor calculation device calculates an anomaly factor based on the reference structure corresponding to an operating state of the target plant or facility learned by the learning device. Fig. Figure 19 is a block diagram showing a configuration example of the anomaly factor calculation device including a corresponding reference structure correction unit in the first embodiment. Fig. Figure 20 is a representation describing a concept of a processing example in which a reference structure correction unit corrects the reference structure or related structure in the anomaly factor calculation device including the reference structure correction unit in the first embodiment. Fig. Figure 21 is a block diagram that shows a configuration example of the anomaly factor calculation device including a relationship change calculation unit in the first embodiment. Fig. Figure 22 is a representation describing a concept of an example of a relationship change sequence calculation processing performed by the relationship change calculation unit on the basis of a learned reference structure and a reference structure at the time of an anomaly in a case where the anomaly factor calculation device according to the first embodiment includes the relationship change calculation unit. Fig. Figure 23 is a block diagram showing a configuration example of the anomaly factor calculation device, which includes an anomaly factor device calculation unit and has a configuration for calculating an anomaly factor in units of the device in the first embodiment. Fig. Figure 24 is a representation describing a concept of an example of a device unit anomaly factor calculation processing for calculating a factor of an anomaly in units of the device, wherein the processing by the anomaly factor device calculation unit is performed on the basis of information from sensors attached to the device, the anomaly detection sequence calculation result and the anomaly propagation sequence calculation result in a case where the anomaly factor calculation device according to the first embodiment includes the anomaly factor device calculation unit. Fig. Figure 25 is a representation showing a screen example of an anomaly factor device calculation result screen displayed on a display device by an anomaly factor calculation result output unit in the first embodiment. Fig. Figure 26 is a block diagram illustrating a configuration example of an anomaly factor calculation device which includes a reference structure diagram output unit and is set up to output reference structure diagram display information to a display device in the first embodiment. Fig. Figure 27 is a representation describing an example of a diagram screen displayed on the display device by the reference structure diagram output unit, which outputs reference structure diagram display information in a case where the anomaly factor calculation device includes the reference structure diagram output unit in the first embodiment. Fig. Figure 28 is a representation that shows an example of the content of the reference structure. Fig. Figure 29 is a representation showing an example of the content of anomaly detection sensor information. Fig. Figure 30 is a representation that shows an example of the content of a result of an anomaly factor order calculation result. Fig. Figure 31 is a flowchart to describe an example of the operation of the anomaly factor calculation device in a case in which the anomaly factor calculation device includes a reference structure diagram output unit in the first embodiment. Fig. Figure 32 is a block diagram showing a configuration example of the learning device with a learning sensor pair generation unit in the first embodiment. Fig. Figure 33 is a representation describing a concept of an example of a method in which the learning sensor pair generation unit generates sensor pair information based on plant or facility design information in a case in which the learning device includes the learning sensor pair generation unit in the first embodiment. DESCRIPTION OF THE EMBODIMENTS

[0012] For a more detailed description of the present disclosure, an embodiment of the present disclosure is described below with reference to the accompanying drawings. First embodiment

[0013] The anomaly factor calculation device according to a first embodiment is used for all plants or facilities in which some type of anomaly occurs in the sensor data collected in the plants, such as in a power plant or a factory automation (FA) system. Furthermore, the sensor data is acquired from sensors present in a plurality of elements (hereinafter referred to as "plant components") that constitute the plant or facility. In the first embodiment, it is assumed that the plant components are, for example, devices. A device is equipped with one or more sensors 300. For the sake of simplicity, in the following first embodiment, it is assumed as an example that one sensor is present in a device.

[0014] For example, the anomaly factor calculation device monitors sensor data collected in a facility (hereinafter referred to as the "target facility" or "target plant") that is a monitoring target, i.e., a target for detecting the occurrence of an anomaly, and monitors a plurality of sensors (hereinafter referred to as "anomaly detection sensors") in which the occurrence of the anomaly has been detected based on the sensor data. Furthermore, as described above, the plant components that constitute the plant, here referred to as devices, interact with one another, and the data collected by the sensors present in and attached to the devices are also of some significance.If an anomaly occurs in a specific plant component out of the majority of plant components and the anomaly is detected, the influence of the anomaly spreads and an anomaly is detected in a multitude of sensor data, in other words, a plurality of sensors.

[0015] When multiple anomaly detection sensors are detected, the anomaly factor calculator calculates or estimates an anomaly factor based on sensor data acquired by these sensors and presents information regarding the calculation result to an operator, such as a site maintenance worker at the target facility. For example, the anomaly factor calculator displays information about the result of the anomaly factor calculation to the operator by prompting a display device to show the information. The anomaly factor calculator presents the operator with the information about the result of the calculation or estimate of the anomaly factor in a format that allows, for example, the identification of a sensor in a device that caused the anomaly or a sequence in which the operator should perform an inspection.In this way, the anomaly factor calculation device can reduce unnecessary inspection work by the operator and decrease the operator's workload. Furthermore, the anomaly factor calculation device can estimate the anomaly factor using a quantitative index that is independent of human subjectivity and provide reasons for the calculation.

[0016] Fig. Figure 1 is a block diagram showing an example of the configuration of a precise diagnostic system 1000 with an anomaly factor calculation device 100 according to the first embodiment.

[0017] The precise diagnostic system 1000 comprises an anomaly factor calculation device 100, a learning device 200, a sensor 300, and a display device 400. It should be noted that in the first embodiment, the sensor 300 and the display device 400 are included in the precise diagnostic system 1000, but this is only an example. The precise diagnostic system 1000 need not necessarily include the sensor 300 and the display device 400, and the sensor 300 and the display device 400 may be contained in a system connected to the precise diagnostic system 1000 but separate from the precise diagnostic system 1000.

[0018] Furthermore, in Fig. For simplicity, only one sensor 300 is shown, but there may be multiple sensors 300. The anomaly factor calculation device 100 is connected to the multiple sensors 300. There may also be multiple display devices 400.

[0019] The anomaly factor calculation device 100 is connected to the learning device 200, the sensor 300, and the display device 400. The anomaly factor calculation device 100 estimates or calculates the factor of an anomaly that has occurred in the target system (not shown).

[0020] In particular, the anomaly factor calculation device 100 detects the sensor 300 in which an anomaly has occurred (hereinafter referred to as the "anomaly detection sensor") on the basis of sensor data acquired by the sensor 300 and a learned associated structure or reference structure generated by the learning device 200, and tracks a propagation path of the anomaly between anomaly detection sensors, thereby calculating the factor of the anomaly that has occurred in the target plant.

[0021] In the first embodiment, the anomaly factor calculation device 100 “calculates or estimates a factor of an anomaly” means that an anomaly factor value, which indicates the degree of probability of the source of the anomaly, and an anomaly factor order based on the anomaly factor value in units of sensors 300 are estimated or calculated, and information regarding the anomaly factor value and the anomaly factor order is generated.

[0022] Then the anomaly factor calculation device 100 causes the display device 400 to display information about the calculated factor of the anomaly.

[0023] Details about the anomaly factor calculation device 100 and its corresponding setup will be described later.

[0024] The learning device 200 determines the reference structure based on sensor data acquired by the sensor 300 present in the target plant during normal operation of the target plant. In the first embodiment, the determination of the reference structure performed by the learning device 200 is also referred to as "learning." That is, a "learned reference structure," which is used when the anomaly factor calculation device 100 calculates the factor of the anomaly that occurred in the target plant, can be referred to as a "calculated structure," which, in other words, is a reference structure estimated by the learning device 200. The reference structure is information that specifies a dependency relationship between a plurality of plant components that constitute the target plant. The reference structure indicates the dependency relationship of the plant components by specifying a dependency relationship between the sensors present in the plant components.The reference structure is, for example, information in which the dependency relationships of the multiple plant components comprising the target plant are represented by a matrix. The reference structure is, for example, information in which the dependency relationships between the multiple components of the plant comprising the target plant are specified in a JavaScript (registered trademark) Object Notification (JSON) format, which includes a dictionary-type description method. In the first embodiment, the reference structure that forms the target plant is, for example, information in which the dependency relationships of a plurality of plant components are specified by a matrix.

[0025] Furthermore, the point in time of normal operation of the target system is specifically a point in time of normal operation of the majority of system components that constitute the target system, in this case, the devices. Thus, the sensor data acquired by sensor 300 during the period of normal operation of the target system are specifically sensor data acquired by sensor 300, which is provided in each device during the period of normal operation of a majority of devices that constitute the target system.

[0026] Details of the learning device 200 will be described later.

[0027] The Sensor 300 is provided in the majority of the system components that make up the target system.

[0028] The sensor 300 forwards the sensor data to the anomaly factor calculation device 100.

[0029] The sensor data are, for example, time-series data of sensor measurements for a predetermined time, acquired at predetermined intervals by the sensor 300, which is present in each device that is a component of the target plant. The sensor data specify, for example, a sensor measurement of at least one degree of opening, deviation, rotational speed, conductivity, flow rate, pressure, temperature, concentration, or water level. Furthermore, this is only one example, and the sensor data can include a control value, such as a command value or a reference value, for a predetermined time, which is obtained at predetermined intervals by the majority of the sensors 300.

[0030] In the following first embodiment, it is assumed that the sensor data are time series data of at least one of the sensor measurements such as an opening degree, a deviation, a rotational speed, a conductivity, a flow rate, a pressure, a temperature, a concentration and a water level for a predetermined time, which are obtained at predetermined intervals from the majority of the sensors 300.

[0031] The display device 400 is, for example, a display contained in a personal computer (PC) installed in an office or similar location where the operator performs work. The display device 400 is, for example, a touch panel display of a tablet terminal carried by the operator.

[0032] First, an anomaly factor calculation device 100 according to the first embodiment is described.

[0033] Fig. Figure 2 is a block diagram showing a configuration example of the anomaly factor calculation device 100 according to the first embodiment.

[0034] It should be noted that the learning device 200 in Fig. 2 is not shown.

[0035] Similar to Fig. 1 is in Fig. For the sake of simplicity, only one sensor 300 is shown in Figure 2, which is merely an example. A plurality of sensors 300 can be connected to the anomaly factor calculation device 100. In the first embodiment, it is assumed that a plurality of sensors 300 are connected to the anomaly factor calculation device 100. In the following first embodiment, the plurality of sensors 300 are also simply referred to as sensor 300.

[0036] The anomaly factor calculation device 100 comprises a sensor data acquisition unit 10, a data storage unit 20, an anomaly detection unit 30, an anomaly detection sequence calculation unit 40, an anomaly propagation path tracking unit 50, an anomaly factor calculation unit 60 and an anomaly factor calculation result output unit 70.

[0037] The sensor data acquisition unit 10 acquires sensor data from sensor 300.

[0038] As described above, in the first embodiment the sensor data are time series data of sensor measurements (e.g. sensor measurements of at least an opening degree, a deviation, a rotational speed, a conductivity, a flow rate, a pressure, a temperature, a concentration or a water level) for a predetermined time, which are obtained at predetermined intervals from the sensors 300, which are provided as system components in the majority of devices.

[0039] For example, if the number of sensors is 300, represented by n, then the 300 sensors are represented by X1, X2, X3, X4, ..., Xn. Assuming that sensor data is acquired at each of the time points 1, 2, ..., t, the sensor data is represented by a two-dimensional data frame, in which one row represents the number t of time points and one column represents the number n of sensors. The sensor data at time 1 of the first sensor X1 is represented by X11, the sensor data at time 1 of the second sensor X2 is represented by X21, and the sensor data at time 2 of the first sensor X1 is represented by X12 (see Fig. 3) In the following first embodiment, the sensor data acquired by the sensor data acquisition unit 10 are referred to as “sensor data D1”.

[0040] The sensor data acquisition unit 10 initiates the storage of the acquired sensor data D1 in the data storage unit 20.

[0041] The anomaly detection unit 30 performs anomaly detection processing on the sensor data D1, which is stored by the sensor data acquisition unit 10 in the data storage unit 20. Specifically, the anomaly detection unit 30 performs the anomaly detection processing on the sensor data D1, which is time-series data stored by the sensor data acquisition unit 10 in the data storage unit 20, using a known univariate anomaly detection method, and detects a plurality of anomaly detection sensors among the sensors 300. In the first embodiment, the occurrence of an anomaly in sensor 300 means that a value of the sensor data acquired by sensor 300 is abnormal. That is, the anomaly detection sensor is the sensor 300 for which a value of the sensor data acquired by sensor 300 is abnormal.It should be noted that in the first embodiment, the occurrence of an anomaly in sensor 300 means that an anomaly has occurred in a device in which sensor 300 is provided.

[0042] Examples of well-known univariate methods for detecting anomalies include Discord (Non-Patent Literature: KEOGH, Eamonn; LIN, Jessica; FU, Ada. Hot Wire: Efficiently finding the most unusual subsequence of a time series. In: Data Mining, fifth international IEEE conference on. IEEE, 2005) and Hotelling's T^2 theory.

[0043] In the following first embodiment, the multiple anomaly detection sensors detected by the anomaly detection unit 30 are also simply referred to as "anomaly detection sensors".

[0044] The anomaly detection unit 30 causes information about the anomaly detection sensor (hereinafter referred to as "anomaly detection sensor information") and information about the detection time at which the anomaly detection sensor was detected (hereinafter referred to as "anomaly detection time information") to be stored in the data storage unit 20.

[0045] In the following first embodiment, the sensor information for anomaly detection is referred to as "Anomaly Detection Sensor Information D3" and the time information for anomaly detection as "Anomaly Detection Time Information D4". The Anomaly Detection Sensor Information D3 is information that identifies the anomaly detection sensor. This information includes, for example, data that can specify the anomaly detection sensor, such as an ID assigned to each sensor. The Anomaly Detection Time Information D4 is information that combines data that can specify the anomaly detection sensor with the time at which the anomaly detection sensor is detected.

[0046] Furthermore, the anomaly detection unit 30 can combine the anomaly detection sensor information D3 and the anomaly detection time information D4 into information (hereinafter referred to as the "anomaly detection result") in which the information specifying the anomaly detection sensor is linked to the time at which the anomaly detection sensor is detected. In this case, the anomaly detection unit 30 causes the anomaly detection result to be stored in the data storage unit 20.

[0047] This assumes that the anomaly detection unit 30 acquires the sensor data D1 from the sensor data acquisition unit 10 via the data storage unit 20, but this is only an example. The anomaly detection unit 30 can acquire the sensor data D1 directly from the sensor data acquisition unit 10.

[0048] The anomaly detection sequence calculation unit 40 acquires the anomaly detection sensor information D3 and the anomaly detection time information D4, which are stored by the anomaly detection unit 30 in the data storage unit 20, and performs an anomaly detection sequence calculation processing to calculate an order in which the occurrence of an anomaly in the anomaly detection sensor was detected, more precisely, an order in which the occurrence of an anomaly in the sensor data D1 collected by the anomaly detection sensor was detected (hereinafter referred to as the "anomaly detection sequence").

[0049] In particular, the anomaly detection sequence calculation unit 40 assigns the anomaly detection sequence to the anomaly detection sensors in the order of the earliest anomaly detection time, based on the anomaly detection sensor information D3 and the anomaly detection time information D4.

[0050] In particular, the anomaly detection sequence calculation unit 40 assigns an anomaly detection sequence to the sensor Xn that has been identified as an anomaly detection sensor. Here, the anomaly detection sequence is a real number. For example, the anomaly detection sequence calculation unit 40 assigns the anomaly detection sequence on such that the assigned anomaly detection sequence on is in ascending order, starting with the sensor Xn with the earliest anomaly detection time. For example, if there are multiple sensors Xn that are assigned to the same anomaly detection time, the anomaly detection sequence calculation unit 40 assigns the same anomaly detection sequence on to the multiple sensors Xn that are assigned to the same anomaly detection time.For example, the anomaly detection sequence calculation unit 40 can assign the anomaly detection sequence “0” to the sensor Xn that is associated with the earliest anomaly detection time, and then assign the elapsed time from the time corresponding to the anomaly detection sequence “0” to the other sensors Xn as anomaly detection sequence on.

[0051] The anomaly detection sequence calculation unit 40 causes a result of the anomaly detection sequence assignment (hereinafter referred to as the "anomaly detection sequence calculation result") to be stored in the data storage unit 20. In the following first embodiment, the anomaly detection sequence calculation result is referred to as "anomaly detection sequence calculation result D5". The anomaly detection sequence calculation result D5 is a piece of information in which the information specifying the anomaly detection sensor, the anomaly detection time, and the information specifying the anomaly detection sequence are mapped to each other.

[0052] It should be noted that this assumes the anomaly detection sequence calculation unit 40 receives the anomaly detection sensor information D3 and the anomaly detection time information D4 from the anomaly detection unit 30 via the data storage unit 20, but this is only an example. The anomaly detection sequence calculation unit 40 can also receive the anomaly detection sensor information D3 and the anomaly detection time information D4 directly from the anomaly detection unit 30.

[0053] The anomaly propagation path tracking unit 50 acquires the anomaly detection sensor information D3, stored by the anomaly detection unit 30, and the reference structure from the data storage unit 20, and performs an anomaly propagation sequence estimation or calculation to determine the propagation sequence of the anomaly (hereinafter referred to as the "anomaly propagation sequence") based on the acquired anomaly detection sensor information D3 and the reference structure. During the anomaly propagation sequence calculation processing, the anomaly propagation path tracking unit 50 calculates the anomaly propagation sequence with respect to sensor 300.It should be noted that for a given sensor 300, the anomaly detection sequence assigned by the anomaly detection sequence calculation unit 40 and the anomaly propagation sequence calculated by the anomaly propagation path tracking unit 50 are not necessarily the same order. For example, the order relationship of the anomaly propagation sequence may reverse due to a time resolution issue, even if the same sequence is assigned to the anomaly detection sequence. Furthermore, for example, due to anomaly detection time accuracy issue, the anomaly detection sequence may be assigned an order opposite to the anomaly propagation sequence. Additionally, for example,Due to an issue with the accuracy of anomaly detection time, even in a case where the anomaly detection sequence is assigned the same order, the anomaly propagation sequence can be reversed. While it is relatively easy to determine the presence or absence of an anomaly, it is difficult to pinpoint the exact time of its occurrence, so a problem with the accuracy of the anomaly detection time can arise.

[0054] The reference structure is generated by learning in the learning device 200 based on at least one statistic from the multitude of sensor data and stored in the data storage unit 20. In the following first embodiment, the reference structure is referred to as "reference structure D2". As described above, in the first embodiment, reference structure D2 consists of information in which the dependency relationship between the plurality of devices, i.e., a plurality of plant components that constitute the target plant, is specified by a matrix.

[0055] In particular, the anomaly propagation path tracking unit 50 determines the dependency relationship between the anomaly detection sensors based on the anomaly detection sensor information D3 and the reference structure D2, and tracks the propagation path of the anomaly by tracing the direction of the dependency relationship. The anomaly propagation path tracking unit 50 then sequentially assigns the anomaly propagation sequence, such as "1" (st), "2" (nd), "3" (rd), ..., from the anomaly detection sensor located upstream of the anomaly propagation and a potential source of the anomaly to the anomaly detection sensor located downstream. The anomaly propagation path tracking unit 50 generates a calculation result of the anomaly propagation sequence (hereinafter referred to as the "anomaly propagation sequence calculation result") and causes the calculation result to be stored in the data storage unit 20.In the following first embodiment, the anomaly propagation sequence calculation result is referred to as "anomaly propagation sequence calculation result D6".

[0056] The following describes the anomaly propagation sequence processing performed by the anomaly propagation path tracking unit 50 using a specific example and the drawings.

[0057] Fig. 4, Fig. 5 and Fig. Figure 6 is a representation describing a specific example of the anomaly propagation sequence calculation processing performed by the anomaly propagation path tracking unit 50 in the first embodiment. <Erzeugung einer Einflussausbreitungsbeziehungsmatrix>

[0058] During the anomaly propagation sequence calculation, the anomaly propagation path tracking unit 50 first converts the reference structure D2 into an influence propagation relationship matrix. This matrix indicates the presence or absence of a dependency relationship between sensor data by means of a matrix based on the reference structure D2 stored in the data storage unit 20. In the first embodiment, the influence propagation relationship matrix is ​​referred to as "influence propagation relationship matrix D9".

[0059] Fig. Figure 4 is a representation to describe a concept of a processing example in which the anomaly propagation path tracking unit 50 transforms the reference structure D2 into the influence propagation relationship matrix D9 in the first embodiment.

[0060] It is assumed that the reference structure D2 is a three-dimensional field in which the first dimension is the type m of the statistical index, the second dimension is the number n of sensors, and the third dimension is the number n of sensors. Furthermore, the statistical index is an index that describes a dependency relationship between the individual parts of the sensor data. The details of the statistical index will be described later.

[0061] A two-dimensional matrix corresponding to the k-th statistical index is denoted A(k), and a statistic describing a dependency relationship between the i-th sensor data and the j-th sensor data in the k-th statistical index is denoted a(k)ij. Note that the i-th sensor data are sensor data acquired by the i-th sensor Xi, and the j-th sensor data are sensor data acquired by the j-th sensor Xj.

[0062] The anomaly propagation path tracking unit 50 performs preprocessing of the selection of statistics (hereinafter referred to as "tracking statistics") to be used for tracking an anomaly propagation path for the reference structure D2, as described above, and generates a reference structure after preprocessing. In the following first embodiment, the reference structure after preprocessing is referred to as "preprocessed reference structure D8 after preprocessing." The preprocessed reference structure D8 after preprocessing has the same data structure as reference structure D2.

[0063] Thus, the anomaly propagation path tracking unit 50 only needs to select a statistic with a large dependency relationship as the tracking statistic. In this case, the anomaly propagation path tracking unit 50 provides a threshold (hereafter referred to as the "statistic selection threshold") for each type of statistical index and selects the statistic a(k)ij with an absolute value |a(k)ij| of a statistic that is greater than the statistical selection threshold as the tracking statistic.For example, if the absolute value of the statistic is greater than the statistic selection threshold, the anomaly propagation path tracking unit 50 replaces the statistic a(k)ij, which is an element of reference structure D2, with an element b(k)ij of the preprocessed reference structure D8 after preprocessing. If the absolute value of the statistic is less than the statistic selection threshold, the anomaly propagation path tracking unit 50 replaces "0", indicating that there is no dependency relationship, in the element b(k)ij of the preprocessed reference structure D8 after preprocessing. Furthermore, the statistic selection threshold can be preset manually by the operator or similar person using an input device (not shown) such as a mouse or keyboard, or it can be automatically determined by the anomaly propagation path tracking unit 50 based on sensor data.For example, the anomaly propagation path tracking unit 50 can determine a relative statistical selection threshold from the mean (median, quantile, etc.) of the statistics.

[0064] The anomaly propagation path tracking unit 50 performs a transformation processing in which the preprocessed reference structure D8 is transformed into the influence propagation relationship matrix D9 after preprocessing in order to generate the influence propagation relationship matrix D9.

[0065] The anomaly propagation path tracking unit 50 determines the dependency relationship between parts of sensor data based on one or more types of statistical indices, including, for example, at least one directed statistical index, and transforms the preprocessed reference structure D8 into the influence propagation relationship matrix D9 after preprocessing. In this case, as a method for determining the dependency relationship between parts of sensor data, the anomaly propagation path tracking unit 50 can, for example, determine that the dependency relationship between the parts of sensor data exists if at least one type of statistic among the statistics b(1)ij, b(2)ij, ..., b(m)ij is not "0", or it can determine that the dependency relationship between the sensor data exists if all statistics b(1)ij, b(2)ij, ..., b(m)ij are not "0".Specifically, assuming i = 1 and j = 2, it can be determined that the dependency relationship between the sensor data collected by sensor X1 and the sensor data collected by sensor X2 exists if at least one type of statistic among the statistics b(1)12, b(2)12, ..., b(m)12 is not "0", or it can be determined that the dependency relationship between the sensor data captured by sensor X1 and the sensor data captured by sensor X2 exists if all statistical indices b(1)12, b(2)12, ..., b(m)12 are not "0".

[0066] The statistical index for the anomaly propagation path tracking unit 50, used to determine the dependency relationship between sensor data, can be manually selected by the operator or similar from m types of statistical indices. As a specific example, the anomaly propagation path tracking unit 50 causes the display device 400 to display a settings screen for the type of statistical index, showing a checkbox or similar for each type of statistical index. The operator or similar uses the input device, such as a mouse or keyboard, to select a statistical index from the settings screen. The anomaly propagation path tracking unit 50 receives the statistical index selected by the operator or similar as the statistical index for determining the dependency relationship between the parts of the sensor data.

[0067] In the first embodiment, the influence propagation relationship matrix D9 is a two-dimensional matrix, where the first dimension is the number n of sensors 300 and the second dimension is the number n of sensors 300, as shown in Fig. 4 shown.

[0068] Here, the elements of the reference structure D2 and the preprocessed reference structure D8 are real numbers after preprocessing, and the elements of the influence propagation relationship matrix D9 are Boolean values. The larger the absolute values ​​| a(k)ij | and | b(k)ij | of the elements in the reference structure D2 and the preprocessed reference structure D8 are after preprocessing, the stronger the dependency relationship, and if the element cij in the influence propagation relationship matrix D9 has the value "1", the existence of the dependency relationship is indicated.

[0069] For example, the anomaly propagation path tracking unit 50 replaces cij, which is an element of the influence propagation relationship matrix D9, with "1" if there is a dependency relationship between the i-th sensor data and the j-th sensor data, and replaces cij, which is an element of the influence propagation relationship matrix D9, with "0" if there is no dependency relationship between the i-th sensor data and the j-th sensor data. <anomalieausbreitungsreihenfolgeberechnung>

[0070] When the anomaly propagation sequence calculation processing generates the influence propagation relationship matrix D9, the anomaly propagation path tracking unit 50 calculates the anomaly propagation sequence based on the anomaly detection sensor information D3 acquired by the data storage unit 20 and the generated influence propagation relationship matrix D9.

[0071] Fig. 5 and Fig. Figure 6 shows a concept for processing in which the anomaly propagation path tracking unit 50 calculates the anomaly propagation sequence based on the anomaly detection sensor information D3 and the influence propagation relationship matrix D9 in the first embodiment.

[0072] In this example, the number of sensors 300 is set to six, and sensor 300 is represented by sensor Xn (n = 1, ..., 6). Furthermore, it is assumed, for example, that sensors X1, X2, X4, and X6 are anomaly detection sensors. Additionally, the influence propagation relationship matrix D9 is a two-dimensional 6 × 6 matrix representing the dependency relationship between the parts of the sensor data with respect to sensor Xn.

[0073] First, the anomaly propagation path tracking unit 50 transforms the influence propagation relationship matrix D9 into an influence propagation graph or influence propagation diagram D10. As in Fig. As shown in Figure 5, the influence propagation graph D10 is a directed graph in which sensors X1, X2, X3, X4, X5, and X6 are represented as nodes, and dependency relationships between the sensor data associated with sensors X1, X2, X3, X4, X5, and X6 are represented as edges. Sensors X1, X2, X3, X4, X5, and X6 correspond to nodes N51, N52, N53, N54, N55, and N56, respectively. For example, if a dependency relationship exists in one direction from sensor X2 to sensor X1, the dependency relationship is expressed by an edge of a one-sided arrow from node N52 to node N51 in the influence propagation graph D10.

[0074] After converting the influence propagation relationship matrix D9 into the influence propagation graph D10, the anomaly propagation path tracking unit 50 then converts the influence propagation graph D10 into an anomaly propagation graph D11 based on the anomaly detection sensor information D3.

[0075] As in Fig. As shown in Figure 5, the anomaly propagation path tracking unit 50, for example, selects only the dependency relationship related to the anomaly detection sensor Xn from the dependency relationship represented by the influence propagation graph D10 and transforms the selected dependency relationship into the anomaly propagation graph D11. In this case, the anomaly propagation path tracking unit 50 selects only the nodes corresponding to the anomaly detection sensors X1, X2, X4, and X6, and the edges between the nodes corresponding to the anomaly detection sensors X1, X2, X4, and X6, from the nodes corresponding to the sensors X1, X2, X3, X4, X5, and X6 represented by the influence propagation graph D10, and the edges between the nodes corresponding to the sensors X1, X2, X3, X4, X5, and X6. In the Fig. In the example shown in Figure 5, sensor X3 is not included in the anomaly detection sensors X1, X2, X4, and X6. Therefore, the anomaly propagation path tracking unit 50 does not select node N53, which corresponds to sensor X3. As a result, node N53, which corresponds to sensor X3, is deleted from the anomaly propagation graph D11, and the edge between node N53 and the node N54 connected to node N53 is also deleted.

[0076] Then the anomaly propagation path tracking unit 50 estimates or determines the anomaly propagation order based on the anomaly propagation graph D11.

[0077] For example, the anomaly propagation path tracking unit 50 assigns the anomaly propagation sequence to the anomaly detection sensor Xn. Here, the anomaly propagation sequence is a real number. The anomaly propagation path tracking unit 50 assigns the anomaly propagation sequence, for example, such that the anomaly propagation sequence is in ascending order, starting with the anomaly detection sensor Xn that is upstream of the anomaly propagation path and is a possible source of the anomaly.

[0078] In the Fig. In the example shown in Figure 5, the anomaly propagation path tracking unit 50 first designates a node to which a one-sided arrow is not drawn from any other node as the node furthest upstream of the anomaly propagation and assigns the lowest anomaly propagation sequence to the anomaly detection sensor Xn corresponding to the node. Fig. The nodes N52, N54, and N56 are not drawn with a one-way arrow from the other nodes. Thus, the anomaly propagation path tracking unit 50 assigns the propagation sequence o2, o4, and o6 to the anomaly detection sensors X2, X4, and X6, which correspond to nodes N52, N54, and N56, respectively. At this point, o2 = o4 = o6. The anomaly propagation path tracking unit 50 sets, for example, o2 = o4 = o6 = 1. That is, the anomaly propagation path tracking unit 50 sets the anomaly propagation sequence of the anomaly detection sensors X2, X4, and X6 to "1" (st).

[0079] Next, the anomaly propagation path tracking unit 50 assigns the anomaly propagation sequence greater than the assigned anomaly propagation sequence to the anomaly detection sensor Xn, which corresponds to the node at the endpoint of the one-sided arrow emanating from the node assigned the smallest anomaly propagation sequence (here, "1"). Furthermore, the anomaly propagation path tracking unit 50 assigns to the anomaly detection sensor Xn the anomaly propagation sequence corresponding to the node at the endpoint of a two-sided arrow emanating from the node assigned the smallest anomaly propagation sequence (here, "1"). Fig. In section 5, node N51 is a node at the endpoint of the one-sided arrow originating from node N52. Therefore, the anomaly propagation path tracking unit 50 assigns the propagation sequence o1 to the anomaly detection sensor X1 corresponding to node N51, i.e., the anomaly propagation sequence o1 greater than "1" (st). For example, the anomaly propagation path tracking unit 50 sets the anomaly propagation sequence o1 = 2. That is, the anomaly propagation path tracking unit 50 sets the anomaly propagation sequence of the anomaly detection sensor X1 to "2" (nd). Furthermore, in Fig. 5 of the nodes N56 at the endpoint of the double arrow originating from node N52, but the anomaly propagation sequence o6 = 1 has already been assigned to node N56.

[0080] Then the anomaly propagation path tracking unit 50 repeats the assignment of the anomaly propagation sequence on as described above until the anomaly propagation sequence on is assigned to the anomaly detection sensors Xn corresponding to all nodes of the anomaly propagation graph D11.

[0081] Furthermore, there can be multiple paths for tracking anomaly propagation, i.e., multiple anomaly propagation sequences to be assigned, depending on how the nodes to be tracked are selected. In the Fig. In the example shown in Figure 5, it is assumed, for instance, that the two-way arrow between node N52 and node N56 is a one-way arrow from node N56 to node N52. In this case, there are two candidates for the anomaly propagation sequence o1 assigned to the anomaly detection sensor X1 corresponding to node N51. Specifically, the anomaly propagation sequence o1 assigned based on the path from node N54 directly to node N51 and the anomaly propagation sequence o1 assigned based on the path from node N56 to node N51 via node N52 are listed as candidates for the anomaly propagation sequence o1 assigned to the anomaly detection sensor X1.In this case, for example, the anomaly propagation path tracking unit 50 assigns a candidate with a later order among the candidates of anomaly propagation sequence o1 to the anomaly propagation sequence o1. Furthermore, this is only an example, and the anomaly propagation path tracking unit 50 could, for example, assign an anomaly propagation sequence o1 of node N51 with an earlier order.

[0082] The anomaly propagation path tracking unit 50 generates a calculation or determination result (hereinafter referred to as the "anomaly propagation sequence calculation result") of the anomaly propagation sequence and causes the calculation result to be stored in the data storage unit 20. In the following first embodiment, the anomaly propagation sequence calculation result is referred to as "anomaly propagation sequence calculation result D6".

[0083] The anomaly propagation sequence calculation result D6, for example, is information in which the information specifying the anomaly detection sensor Xn (specified by D6A in Fig. 5), an anomaly detection sensor flag fn (specified by D6B in Fig. 5), which indicates whether the anomaly detection sensor Xn is included in the anomaly detection sensor or not, and the anomaly propagation sequence on (specified by D6C in Fig. 5), which are determined by the anomaly propagation path tracking unit 50, are assigned to each other. It should be noted that in the Fig. In the example shown in Figure 5, only the anomaly detection sensor Xn is included as sensor Xn, which is included in the anomaly propagation sequence calculation result D6. The anomaly detection sensor flag fn is a Boolean value. Since in the Fig. In the example shown in Figure 5, all anomaly detection sensors Xn that are specified in the anomaly propagation sequence calculation result D6, in particular the anomaly detection sensors X1, X2, X4 and X6, are anomaly detection sensors, the anomaly propagation path tracking unit 50 assigns, for example, the anomaly detection sensor flag fn of the anomaly detection sensors X1, X2, X4 and X6 (true).

[0084] It should be noted that when determining the anomaly propagation sequence based on the anomaly detection sensor information D3 and the influence propagation relationship matrix D9 by the anomaly propagation path tracking unit 50, which refers to Fig. As described in section 5, the anomaly propagation path tracking unit 50 only selects the dependency relationship between the parts of the sensor data that relate to the anomaly detection sensor Xn when it transforms the influence propagation graph D10 into the anomaly propagation graph D11, but this is merely an example.

[0085] As in Fig. As shown in Figure 6, when converting the influence propagation graph D10 into the anomaly propagation graph D11, the anomaly propagation path tracking unit 50 can, for example, select a dependency relationship between the parts of sensor data relating to sensor Xn if at least one of two different sensors Xn is the anomaly detection sensor Xn.

[0086] In this case, the anomaly propagation path tracking unit 50 selects an edge where at least one of the connected nodes between nodes N51, N52, N53, N54, N55, and N56, corresponding to sensors X1, X2, X3, X4, X5, and X6 as represented by the influence propagation graph D10, is a node corresponding to anomaly detection sensors X1, X2, X4, and X6, and is connected to the edge. For example, node N53, corresponding to sensor X3, and node N54, corresponding to sensor X4, are connected by the edge of the double-sided arrow. In this case, sensor X3 is not included in anomaly detection sensor Xn, but sensor X4 is included in anomaly detection sensor Xn.In the anomaly propagation graph D11, the edge between node N53 and node N45 is therefore not deleted.

[0087] Furthermore, in this case, for example, the anomaly propagation path tracking unit 50 can generate the anomaly propagation graph D11 such that nodes N53 and N55, corresponding to sensors X3 and X5, which are not included in the anomaly detection sensor Xn, can understand this from nodes N51, N52, N53, N54, and N55, corresponding to sensors X1, X2, X3, X4, X5, and X6, which are represented by the anomaly propagation graph D11. In the Fig. In the anomaly propagation graph or diagram D11 shown in section 6, nodes N51, N52, N54 and N56 are represented by solid circles and nodes N53 and N55 by dotted circles.

[0088] Furthermore, in this case, the anomaly propagation path tracking unit 50 determines the anomaly propagation sequence based on the anomaly propagation graph D11, which contains nodes N53 and N55 corresponding to sensors X3 and X5 of sensors Xn that are not included in the anomaly detection sensor Xn.

[0089] For example, in Fig. 6. Nodes N53 and N54 are not drawn with a one-sided arrow from other nodes. The anomaly propagation path tracking unit 50 therefore assigns the propagation sequence o3 and o4 to sensors X3 and X4, respectively. At this point, o3 = o4. The anomaly propagation path tracking unit 50 sets, for example, o3 = o4 = 1. That is, the anomaly propagation path tracking unit 50 sets the anomaly propagation sequence o3 and o4 of sensors X3 and X4, respectively, to "1" (st).

[0090] Next, the anomaly propagation path tracking unit 50 assigns the anomaly propagation sequences o1 and o5, which are greater than the anomaly propagation sequence o4, to sensors X1 and X5, corresponding to nodes N51 and N55 at the endpoints of the one-sided arrows emanating from node N54, which was assigned the lowest anomaly propagation sequence (here "1"). For example, the anomaly propagation path tracking unit 50 sets o1 = o5 = 2. That is, the anomaly propagation path tracking unit 50n sets the anomaly propagation sequence of sensors X1 and X5 to "2" (nd). Furthermore, the anomaly propagation path tracking unit 50 assigns the anomaly propagation sequences o2 and o6, which are greater than the anomaly propagation sequence o5, to the sensors X2 and X6, which correspond to N52 and N56 respectively at the endpoints of the one-sided arrows emanating from node N55.For example, the anomaly propagation path tracking unit 50 sets o2 = o6 = 3. That is, the anomaly propagation path tracking unit 50 sets the anomaly propagation sequence of sensors X2 and X6 to "3" (rd).

[0091] Furthermore, the anomaly propagation path tracking unit 50 can reassign the anomaly propagation sequence o2, i.e., the sequence greater than "3" (rd), as anomaly propagation sequence o1 to sensor X1, which corresponds to node N51, even though the anomaly propagation sequence o1, i.e., "2" (nd), has already been assigned to X1, which corresponds to node N51, since node N51 is a node at the endpoint of the one-sided arrow emanating from node N52. For example, the anomaly propagation path tracking unit 50 can reassign the anomaly propagation sequence o1 to "4" (th).

[0092] In the Fig. In the example shown in Figure 6, nodes N51, N52, N53, N54, N55, and N56 in the anomaly propagation graph D11 include nodes N53 and N55, which correspond to sensors X3 and X5, respectively, and are not included in the anomaly detection sensor Xn. Therefore, after assigning anomaly propagation sequences o1, o2, o3, o4, o5, and o6 to sensors X1, X2, X3, X4, X5, and X6 respectively, the anomaly propagation path tracking unit 50 can weight the anomaly propagation sequences o3 and o5 assigned to sensors X3 and X5. In the example shown in Figure 6, the anomaly propagation path tracking unit 50 can weight the anomaly propagation sequences o3 and o5 assigned to sensors X3 and X5. Fig. In the example shown in Figure 6, the weight b is added to each of the anomaly propagation sequences o3 and o5, which correspond to the sensors X3 and X5 contained in the anomaly propagation sequence calculation result D6. Furthermore, the weight b is a real number equal to or greater than 0.

[0093] Then the anomaly propagation path tracking unit 50 causes the anomaly propagation sequence calculation result D6 to be stored in the data storage unit 20.

[0094] As described above, the following include in Fig. In the example described in section 6, the sensors Xn, in particular sensors X1, X2, X3, X4, X5 and X6, which are specified in the anomaly propagation sequence calculation result D6, and sensors X3 and X5, which are not included in the anomaly detection sensor Xn (in particular anomaly detection sensors X1, X2, X4 and X6). The anomaly propagation path tracking unit 50 assigns the value (False) to the anomaly detection sensor flag fn of sensors X3 and X5.

[0095] It should be noted that this assumes the anomaly propagation path tracking unit 50 receives the anomaly detection sensor information D3 from the anomaly detection unit 30 via the data storage unit 20, but this is only an example. The anomaly propagation path tracking unit 50 can also receive the anomaly detection sensor information D3 directly from the anomaly detection unit 30.

[0096] The description returns to the one in Fig. 2 shown configuration example of the anomaly factor calculation device 100.

[0097] The anomaly factor calculation unit 60 receives the anomaly detection sequence calculation result D5, which is output by the anomaly detection sequence calculation unit 40, and the anomaly propagation sequence calculation result D6, which is output by the anomaly propagation path tracking unit 50, from the data storage unit 20 and performs an anomaly factor calculation processing to determine a factor of the anomaly based on the anomaly detection sequence determined by the anomaly detection sequence calculation unit 40 and the anomaly propagation sequence determined by the anomaly propagation path tracking unit 50.

[0098] In particular, the anomaly factor calculation unit 60 calculates the anomaly factor value, which indicates the degree of probability of the anomaly's generation source based on the anomaly detection sequence calculation result D5 and the anomaly propagation sequence calculation result D6, and assigns a sequence (hereinafter referred to as the "anomaly factor sequence") based on the calculated anomaly factor value. The higher the degree of probability of an anomaly source, the lower the anomaly factor value.

[0099] First, the anomaly factor calculation unit 60 receives the anomaly detection sequence calculation result D5 and the anomaly propagation sequence calculation result D6 from the data storage unit 20. The anomaly factor calculation unit 60 then calculates a corresponding anomaly factor value for each sensor 300 from the anomaly detection sequence contained in the anomaly detection sequence calculation result D5 and the anomaly propagation sequence contained in the anomaly propagation sequence calculation result D6. Here, the anomaly detection sequence, the anomaly propagation sequence, and the anomaly factor value are all real numbers. In addition, the anomaly factor calculation unit 60 calculates the corresponding anomaly factor value for all sensors 300 that are included in the anomaly detection sequence calculation result D5 or anomaly propagation sequence calculation result D6.

[0100] For example, the Anomaly Factor Calculation Unit 60 calculates a representative value as the anomaly factor using the weighted average of the anomaly detection sequence and the anomaly propagation sequence. The Anomaly Factor Calculation Unit 60 can calculate a representative value, such as a minimum or a maximum, as the anomaly factor. That is, the Anomaly Factor Calculation Unit 60 can, for example, calculate a smaller or larger value for the anomaly detection sequence and the anomaly propagation sequence as the anomaly factor.

[0101] If only the anomaly detection order or the anomaly propagation order is specified, the anomaly factor calculation unit 60 can weight the anomaly factor value, taking into account that only the specified order is set. For example, if only the anomaly detection order or the anomaly propagation order is specified, the anomaly factor calculation unit 60 adds the weight b to the anomaly factor value. Furthermore, the weight b is a real number equal to or greater than 0.For example, if the anomaly propagation path tracking unit 50 assigns the anomaly propagation order under the assumption that the node corresponding to sensor Xn, which is not included in the anomaly detection sensor Xn, is also included when transforming the impact propagation graph D10 into the anomaly propagation graph D11, a situation may arise where the anomaly detection order is not set, but the anomaly propagation order is set.

[0102] After calculating the anomaly factor value for each sensor 300, the anomaly factor calculation unit 60 then assigns an anomaly factor sequence to the sensor 300 based on the calculated anomaly factor value. Furthermore, the anomaly factor sequence is a real number.

[0103] For example, the anomaly factor calculation unit 60 assigns the anomaly factor sequence such that the anomaly factor sequence for sensor 300 is in ascending order, starting with the sensor 300 with the smallest corresponding anomaly factor value. In a case where there are multiple sensors 300 with the same corresponding anomaly factor values, the anomaly factor calculation unit 60 assigns the same anomaly factor sequence to the majority of sensors 300.

[0104] When the anomaly factor sequence is assigned to each sensor 300, the anomaly factor calculation unit 60 generates information regarding the anomaly factor sequence assigned to each sensor 300 (hereinafter referred to as the "anomaly factor sequence calculation result") and causes the information to be stored in the data storage unit 20. In the following first embodiment, the anomaly factor sequence calculation result is referred to as "anomaly factor sequence calculation result D7".

[0105] The anomaly factor sequence calculation result D7 is a piece of information in which information specifying sensor 300, an anomaly detection sensor flag indicating whether sensor 300 is an anomaly detection sensor or not, which is included in the anomaly factor sequence calculation result D5, an anomaly factor value and an anomaly factor sequence are mapped to each other.

[0106] Furthermore, the anomaly detection sensor flag is a Boolean value. For example, the anomaly factor calculation unit 60 assigns (True) to the anomaly detection sensor flag corresponding to sensor 300 if sensor 300 is the anomaly detection sensor, and assigns (False) to the anomaly detection sensor flag corresponding to sensor 300 if sensor 300 is not the anomaly detection sensor. For example, if the information about sensor 300 is contained in the anomaly detection sequence calculation result D5, the anomaly factor calculation unit 60 only needs to determine that sensor 300 is the anomaly detection sensor Xi.

[0107] The anomaly factor calculation processing, which is carried out by the anomaly factor calculation unit 60 as described above, is described with specific examples with reference to the drawings.

[0108] Fig. Figure 7 is a representation to describe a specific example of the anomaly factor calculation processing performed by the anomaly factor calculation unit 60 in the first embodiment.

[0109] In Fig. In section 7, the number of sensors 300 is represented as n (n = 1 to 6), and the number n of sensors 300 is represented as sensor Xn. Furthermore, it is assumed that among the sensors Xn, sensors X1, X2, X4, and X6 are anomaly detection sensors.

[0110] Furthermore, in the Fig. In example 7, the anomaly propagation sequence calculation result D6, which is used by the anomaly factor calculation unit 60 for anomaly factor calculation processing, is information that records only the anomaly propagation sequence corresponding to the anomaly detection sensor Xi, as shown in Fig. 5 shown.

[0111] In the Fig. In the example shown in Figure 7, the anomaly factor calculation unit 60 receives the anomaly detection sequence calculation result D5 and the anomaly propagation sequence calculation result D6 with respect to the anomaly detection sensors X1, X2, X4 and X6 from the data storage unit 20.

[0112] In this case, the anomaly detection sequence calculation result D5 assumes that the anomaly detection sequence is o1, o2, o4 and o6 (specified by D5C in Fig. 7), corresponding to the anomaly detection sensors X1, X2, X4 and X6, are “2”(nd), “3”(rd), “1”(st) and “4”(th), respectively. Furthermore, the anomaly propagation sequence calculation result D6n contains the propagation sequences o1, o2, o4 and o6 (in Fig. 7 marked by D6C), which correspond to the anomaly detection sensors X1, X2, X4 and X6, “2”(nd), “1”(st), “1”(st) and “1”(st).

[0113] This assumes that the anomaly factor calculation unit 60 calculates an anomaly factor value sn (n = 1, 2, 4, 6) corresponding to the anomaly detection sensor Xn using the average. In this case, the anomaly factor calculation unit 60 calculates anomaly factor values ​​s1, s2, s4, and s6, corresponding to the anomaly detection sensors X1, X2, X4, and X6, as "2", "2", "1", and "2.5", respectively. Based on the calculated anomaly factor values ​​s1, s2, s4 and s6, the anomaly factor calculation unit 60 assigns the values ​​“2”(nd), “2”(nd), “1”(st) and “3”(rd) to the anomaly factor sequences o1, o2, o4 and o6, which correspond to the anomaly detection sensors X1, X2, X4 and X6.

[0114] Then the anomaly factor calculation unit 60 generates the anomaly factor sequence calculation result D7.

[0115] In particular, the anomaly factor calculation unit 60 generates the anomaly factor sequence calculation result D7, in which information from the anomaly detection sensors X1, X2, X4 and X6 (displayed by D7A in Fig. 7), Anomaly detection sensor flags f1, f2, f4 and f6 (indicated by D7B in Fig. 7), indicating whether the anomaly detection sensors X1, X2, X4 and X6 are included in the anomaly detection sequence calculation result D5, anomaly factor values ​​s1, s2, s4 and s6 (displayed by D7C in Fig. 7) and anomaly factor sequences o1, o2, o4 and o6 (indicated by D7D in Fig. 7) are assigned to each other.

[0116] In this case, the sensor Xn, which is included in the anomaly detection sequence calculation result D5, is equal to the anomaly detection sensor Xn. Thus, all anomaly detection sensor flags f1, f2, f4, and f6, corresponding to the anomaly detection sensors X1, X2, X4, and X6, are (True).

[0117] Then the anomaly factor calculation unit 60 causes the generated anomaly factor sequence calculation result D7 to be stored in the data storage unit 20.

[0118] It should be noted that this assumes that the anomaly factor calculation unit 60 receives the anomaly detection sequence calculation result D5 from the anomaly detection sequence calculation unit 40 via the data storage unit 20, and the anomaly propagation sequence calculation result D6 from the anomaly propagation path tracking unit 50 via the data storage unit 20; however, this is only an example. The anomaly factor calculation unit 60 can receive the anomaly detection sequence calculation result D5 and the anomaly propagation sequence calculation result D6 directly from the anomaly detection sequence calculation unit 40 and the anomaly propagation path tracking unit 50, respectively.

[0119] The description returns to the one in Fig. 2 shown configuration example of the anomaly factor calculation device 100.

[0120] The anomaly factor calculation result output unit 70 captures the anomaly factor sequence calculation result D7, which is output by the anomaly factor calculation unit 60 from the data storage unit 20, the anomaly detection sequence calculation result D5, which is output by the anomaly detection sequence calculation unit 40, and the anomaly propagation sequence calculation result D6, which is output by the anomaly propagation path tracking unit 50, and outputs information regarding the result of the calculation of the anomaly factor by the anomaly factor calculation unit 60.

[0121] In particular, the anomaly factor calculation result output unit 70, based on the anomaly factor calculation result D7, the anomaly detection sequence calculation result D5, and the anomaly propagation sequence calculation result D6, outputs information (hereinafter referred to as "anomaly factor calculation result display information") to cause the display device 400 to display a screen (hereinafter referred to as the "anomaly factor calculation result screen") that shows information regarding the calculation result of the factor of the anomaly by the anomaly factor calculation unit 60.

[0122] It should be noted that in the first embodiment, the anomaly factor calculation result output unit 70 is provided in the anomaly factor calculation device 100, but this is only an example. The anomaly factor calculation result output unit 70 can be contained in a device (not shown), such as a display, which is connected to the anomaly factor calculation device 100 via a wired or wireless signal line.

[0123] Fig. 8 and Fig. Figure 9 shows examples of screen displays of the anomaly factor calculation result screen, which is displayed on the display device 400 by the anomaly factor calculation result output unit 70 in the first embodiment.

[0124] Fig. 8 and Fig. Figure 9 shows examples of the anomaly factor calculation result screen when the number of sensors is 300 (sensor Xn. n = 1 to 6), as an example.

[0125] In the Fig. 8 and Fig. 9 are the anomaly factor calculation result screen labeled “D12-1” and “D12-2”.

[0126] As in the Fig. 8 and Fig. As shown in Figure 9, the anomaly factor calculation result screen contains, for example, eleven display frames: a display frame D12A, a display frame D12B, a display frame D12C, a display frame D12D, a display frame D12E, a display frame D12F, a display frame D12G, a display frame D12H, a display frame D12l, a display frame D12J and a display frame D12K.

[0127] For example, the anomaly factor calculation result output unit 70 causes a list of anomaly factor calculation results to be displayed on the anomaly factor calculation results screen, containing information about the anomaly factor calculation results. The anomaly factor calculation results list is, for example, a list that displays the information about sensor Xn, the anomaly detection sensor flag, the anomaly detection time, the anomaly detection sequence, the anomaly propagation sequence, the anomaly factor value, and the anomaly factor sequence for each sensor Xn, in relation to each other. In the Fig. 8 and Fig. In the anomaly factor calculation results screen shown in 9, the lists of results of the anomaly factor calculation are labelled with “D12-1a” and “D12-2a”.

[0128] The Fig. The screenshot 8, showing an example of the anomaly factor calculation result screen, is an example of a case where the content of the anomaly detection sequence calculation result D5 is as shown in Fig. As shown in 10A, the content of the anomaly propagation sequence calculation result D6 is as shown in Fig. 10B is shown and the content of the anomaly factor sequence calculation results D7 as in Fig. 10C is shown.

[0129] The anomaly factor calculation result output unit 70 outputs the anomaly factor calculation result display information to the display device 400, causing the information specifying the sensor Xn of the anomaly factor sequence calculation result D7 to be displayed in display frame D12A, causing the information specifying the anomaly detection sensor flag of the anomaly factor sequence calculation result D7 to be displayed in display frame D12B, causing information specifying the anomaly detection time of the anomaly detection sequence calculation result D5 to be displayed in display frame D12C, causing the anomaly detection sequence of the anomaly detection sequence calculation result D5 to be displayed in display frame D12D, causing the anomaly detection sequence of the anomaly propagation sequence calculation result D6 to be displayed in display frame D12E, causingthat the anomaly factor value of the anomaly factor sequence calculation result D7 is displayed in display frame D12F, causes the anomaly factor sequence of the anomaly factor sequence calculation result D7 to be displayed in display frame D12G, causes sort buttons to rearrange the order of the anomaly factor calculation result list in ascending order based on the anomaly detection sequence of the anomaly detection sequence calculation result D5, the anomaly propagation sequence of the anomaly propagation sequence calculation result D6 and the anomaly factor sequence of the anomaly factor sequence calculation result D7 in ascending order in display frames D12l, D12J and D12K respectively, and causes a checkbox to receive an instruction to display only the anomaly detection sensor,displayed in the display frame D12H. As a result, the display device 400 shows the anomaly factor calculation result screen, as shown in , Fig. 8 shown.

[0130] For example, the anomaly factor calculation result output unit 70 causes "True" to be displayed in display frame D12B when (True) is set to the anomaly detection sensor flag of the anomaly factor sequence calculation result D7, and causes "False" to be displayed in display frame D12B when (False) is set to the anomaly detection sensor flag.

[0131] For example, if no value is set for the anomaly detection time or the anomaly detection sequence of the anomaly detection sequence calculation result D5, the anomaly factor calculation result output unit 70 causes a blank character to be displayed.

[0132] Furthermore, the anomaly factor calculation result output unit 70, for example, causes the information specifying sensor Xn, the information specifying the anomaly detection sensor flag, the anomaly detection time, the anomaly detection sequence, the anomaly propagation sequence, the anomaly factor value, and the anomaly factor sequence, which are assigned to sensor Xn according to the sequence of the ID assigned to sensor Xn, to be displayed together in the initial state of the anomaly factor calculation result list. Here, the initial state of the anomaly factor calculation result list refers to the state of the anomaly factor calculation result list when the anomaly factor calculation result output unit 70 causes the display device 400 to display the anomaly factor calculation result list for the first time after power is applied.

[0133] An example of the anomaly factor calculation results list, which is in Fig. Figure 8 is an example of the anomaly factor calculation results list in an initial state.

[0134] In the initial state of the anomaly factor calculation result list, the anomaly factor calculation result output unit 70 sets the anomaly factor calculation result list to a state in which the instruction to rearrange the data and the instruction to display only the anomaly detection sensor are not executed, as in Fig. 8 shown.

[0135] The operator checks, for example, the anomaly factor calculation result screen, as shown in Fig. Figure 8 illustrates this. The operator receives information about the calculation result of the anomaly factor. For example, using the information from sensor Xn that caused the anomaly, the operator can identify the device that caused it. Furthermore, the operator can determine the sequence in which the devices where the anomaly occurred should be inspected. This allows the operator to reduce unnecessary inspection work and decreases the operator's workload.

[0136] If the in Fig. When the anomaly factor calculation results list shown in section 8 is displayed, the operator can also instruct the information displayed in the anomaly factor calculation results list to be rearranged.

[0137] For example, the operator can use an input device (not shown), such as a mouse or keyboard, to press the sort buttons on display frames D12l, D12J, and D12K, thereby instructing the sorting of information. For example, upon receiving an instruction to rearrange the information, the anomaly factor calculation result output unit 70 causes the sort button used to input the instruction—that is, the pressed sort button—to turn black. Then, the anomaly factor calculation result output unit 70 outputs the anomaly factor calculation result display information to display device 400 to show an anomaly factor calculation result list in which the information to be displayed has been rearranged according to the input instruction.In this way, the anomaly factor calculation results list displayed on display device 400 is updated to the anomaly factor calculation results list in which the displayed information is rearranged.

[0138] Furthermore, the operator can enter an instruction to display only the information about the anomaly detection sensor in the anomaly factor calculation results list by operating the input device and pressing the checkbox displayed in display frame D12H. For example, if the instruction to display only the information about the anomaly detection sensor in the anomaly factor calculation results list is entered, the anomaly factor calculation results output unit 70 causes a checkmark to appear in the checkbox. The anomaly factor calculation results output unit 70 then outputs the anomaly factor calculation results display information to the display device 400, causing only the information relating to the anomaly detection sensor to be displayed in the anomaly factor calculation results list according to the input instruction.In this way, the anomaly factor calculation results list displayed on the display device 400 is updated to the anomaly factor calculation results list, which only displays the information about the anomaly detection sensor.

[0139] Fig. Figure 9 is a representation illustrating an example of the anomaly factor calculation results screen, in a state where the anomaly factor calculation results screen appears as shown in Fig. As shown in Figure 8, the sort button in display frame D12K of the anomaly factor calculation result list is pressed by the operator, the information of the anomaly factor calculation result list is reordered in ascending order of the anomaly factor sequence by the anomaly factor calculation result output unit 70, which received the press, the checkbox of display frame D12H is then pressed by the operator, and the anomaly factor calculation result list is displayed, in which only the line corresponding to the anomaly detection sensor whose anomaly detection sensor flag is (True) is displayed by the anomaly factor calculation result output unit 70, which received the press.

[0140] On the Fig. The anomaly factor calculation results screen shown in 9 displays the anomaly factor calculation results list, in which only the rows corresponding to sensors X1, X2, X4 and X6 (i.e. the anomaly detection sensors) with the anomaly detection sensor flag (True) are shown, and the rows corresponding to sensors X1, X2, X4 and X6 are reordered in ascending order of the anomaly factor sequence of the row corresponding to sensor X4, the row corresponding to sensor X2, the row corresponding to sensor X6 and the row corresponding to sensor X1, based on the anomaly factor sequence.

[0141] In the Fig. On the anomaly factor calculation result screen shown in Figure 9, a checkmark is displayed in the checkbox of display frame D12H. This allows the operator to see that only the anomaly detection sensor is displayed on the anomaly factor calculation result screen. Furthermore, in the Fig. In the anomaly factor calculation results screen shown in Figure 9, the sort button in display frame D12K is filled. This allows the operator to see that the lines of the anomaly factor calculation results list are arranged in ascending order of the anomaly factor sequence on the anomaly factor calculation results screen.

[0142] For example, if the operator presses the same sort button again in a state where any sort button among those displayed in display frames D12H, D12J, and D12K is pressed—that is, in a state where the anomaly factor calculation result list display has been reordered—it is possible to revert the anomaly factor calculation result list display to a state prior to the reordering instruction. Upon detecting that the same sort button has been pressed again, the anomaly factor calculation result output unit 70 causes the sort button that was displayed in a filled state to be displayed as empty.Then, the anomaly factor calculation result output unit 70 outputs the anomaly factor calculation result display information to the display device 400 to display the anomaly factor calculation result list before sorting. Therefore, the anomaly factor calculation result list displayed on the display device 400 is updated to the anomaly factor calculation result list before sorting (see ). Fig. 8).

[0143] For example, if the operator presses the checkbox again in a state where the checkbox in display frame D12H is pressed—that is, in a state where the anomaly factor calculation result list, showing only the information about the anomaly detection sensor, is displayed—the display of the anomaly factor calculation result list can be reset to the state before the instruction to display only the information about the anomaly detection sensor. If the checkbox is pressed again, the anomaly factor calculation result output unit 70 causes an unchecked checkbox to be displayed. Then, the anomaly factor calculation result output unit 70 outputs the anomaly factor calculation result display information to the display device 400 to display the anomaly factor calculation result list before switching to displaying only the information relating to the anomaly detection sensor.Therefore, the anomaly factor calculation results list displayed on the display device 400 is transferred to the anomaly factor calculation results list (see . Fig. 8) updated before switching to displaying only the information about the anomaly detection sensor.

[0144] It should be noted that the anomaly factor calculation result screen in its initial state is a screen as shown in Fig. Figure 8 is shown, but this is only an example. For instance, in its initial state, the anomaly factor calculation result output unit 70 on the anomaly factor calculation result screen can either pre-order the information regarding the sensors 300 displayed in the anomaly factor calculation result list in ascending order based on the anomaly detection order, the anomaly propagation order, and the anomaly factor order, or display the anomaly factor calculation result list showing only the anomaly detection sensors.

[0145] The description returns to the one in Fig. 2 shown configuration example of the anomaly factor calculation device 100.

[0146] The data storage unit 20 stores various types of information.

[0147] Specifically, the data storage unit 20 stores, for example, the reference structure D2 generated by the learning device 200, the sensor data D1 acquired by the sensor data creation unit 10, the anomaly detection sensor information D3 and the anomaly detection time information D4 output by the anomaly detection unit 30, the anomaly detection sequence calculation result D5 output by the anomaly detection sequence calculation unit 40, the anomaly propagation sequence calculation result D6 output by the anomaly propagation path tracking unit 50, and the anomaly factor sequence calculation result D7 output by the anomaly factor calculation unit 60.

[0148] It should be noted that in a second embodiment, as in Fig. Figure 2 shows the data storage unit 20 contained within the anomaly factor calculation device 100, but this is only an example. The data storage unit 20 can be located outside the anomaly factor calculation device 100 at a location accessible to the anomaly factor calculation device 100.

[0149] An example of the configuration of the learning device 200 according to the first embodiment is described.

[0150] Fig. Figure 11 is a block diagram showing a configuration example of the learning device 200 according to the first embodiment.

[0151] The learning device 200 performs the learning process using sensor data acquired by sensor 300 in the target system during its normal operation. Specifically, the learning device 200 estimates the reference structure D2 based on this sensor data. It is important to note that the period of normal operation of the target system refers specifically to the period of normal operation of the multiple devices that comprise it. Therefore, the sensor data acquired by sensor 300 during the period of normal operation of the target system is specifically sensor data acquired by the sensor 300 present in each device during the period of normal operation of the multiple devices that constitute the target system.

[0152] The learning device 200 causes the learned reference structure D2 to be stored in the data storage unit 20 of the anomaly factor calculation device 100.

[0153] For the sake of simplicity, in Fig. 11 only the data storage unit 20 is shown as the configuration unit of the anomaly factor calculation device 100.

[0154] The learning device 200 comprises a learning sensor data acquisition unit 210, a learning data storage unit 220, a learning preprocessing unit 230 and a reference structure learning unit 240.

[0155] The learning sensor data acquisition unit 210 acquires training data that is used to learn the reference structure D2. The training data comprises sensor data acquired by the majority of the sensors 300.

[0156] It should be noted that not all of the training data acquired by the training sensor data acquisition unit 210 is used for learning the reference structure D2. The training preprocessing unit 230, which will be described later, acquires the training data that is actually used for learning the reference structure D2, based on the training data acquired by the training sensor data acquisition unit 210. Therefore, the training data acquired by the training sensor data acquisition unit 210 is more accurately described as a candidate training data. The details of the training preprocessing unit 230 will be described later.

[0157] The learning sensor data acquisition unit 210 initiates the storage of the acquired learning data candidate in the learning data storage unit 220.

[0158] The training data candidate is information of the same type as the sensor data D1 acquired by the anomaly factor calculation device 100 from sensor 300, in particular information of the same content (a measured value or a control value of at least one degree of opening, deviation, rotational speed, conductivity, flow rate, pressure, temperature, concentration, or water level) as the sensor data D1, and it is sensor data acquired by sensor 300 during the normal operation of the majority of devices, i.e., the majority of plant components of the target plant. That is, the training data candidate contains a sensor measured value or a control value during the normal operation of the target plant.It should be noted that the learning data candidate is the same type of information as the sensor data D1, which is acquired by the same sensor 300 as the sensor 300, which is the source of the sensor data D1 through the anomaly factor calculation device 100.

[0159] The training data candidate is prepared in advance, for example by an administrator, and stored in a location accessible to the training device 200. The training data candidate can, for instance, be stored in the data storage unit 20 of the anomaly factor calculation device 100. In this case, the training sensor data acquisition unit 210 only needs to retrieve the training data candidate from the data storage unit 20 of the anomaly factor calculation device 100.

[0160] In the first embodiment, the learning data candidate procured by the learning sensor data acquisition unit 210 is designated as "learning data candidate D21".

[0161] It should be noted that in Fig. 11. A learning data candidate obtained from a location other than the anomaly factor calculation device 100 by the learning sensor data acquisition unit 210 is represented as "learning data candidate D21", and a learning data candidate obtained from the anomaly factor calculation device 100 by the learning sensor data acquisition unit 210 is represented as "learning data candidate D22" in a distinguishing manner. This is for the sake of clarity, and the contents of "learning data candidate D21" and "learning data candidate D22" are the same.

[0162] Therefore, in the following description, both the “learning data candidate D21” and the “learning data candidate D22” are referred to as “learning data candidate D21”.

[0163] The learning preprocessing unit 230 takes the learning data candidate D21, acquired by the learning sensor data acquisition unit 210, from the learning data storage unit 220 and performs preprocessing of the learning data candidate D21. It should be noted that the learning preprocessing unit 230 retrieves the learning data candidate D21, stored at a predetermined time, from the learning data storage unit 220 at every predetermined time.

[0164] Specifically, the learning preprocessing unit 230 performs a data conversion, selection or similar operation on the learning data candidate D21 and obtains learning data that is actually used when learning the reference structure D2.

[0165] For example, the learning preprocessing unit 230 converts the training data candidate D21 into a first-order difference series and sets the converted training data candidate D21 as the training data. In this case, the sensor data contained in the training data candidate D21 are converted into data that indicate a change amount.

[0166] Furthermore, the learning preprocessing unit 230 can, for example, select only sensor data that exhibit a large deviation from the sensor data contained in the training data candidate D21 and use the selected sensor data as training data. In this case, the learning preprocessing unit 230 specifies a variance threshold and selects sensor data whose variance is greater than the threshold from the sensor data contained in the training data candidate D21 as training data. It should be noted that the variance threshold is set manually, for example, by the operator or a similar person. The operator or a similar person uses the input device, such as a mouse or keyboard, to input and set the variance threshold.Furthermore, the learning preprocessing unit 230 can, for example, in a case where a measurement error of the sensor 300 is known, use the square of the measurement error as the threshold for the variance.

[0167] The learning preprocessing unit 230 initiates the storage of the acquired learning data in the learning data storage unit 220. In the first embodiment, the learning data acquired by the learning preprocessing unit 230 are designated as "learning data D23".

[0168] In the following first embodiment, it is assumed, for example, that the preprocessing performed on the learning data candidate D21 by the learning preprocessing unit 230 is a selection, and the learning preprocessing unit 230 acquires as the learning data D23 sensor data that have a variance greater than the threshold among the sensor data contained in the learning data candidate D21.

[0169] It should be noted that this assumes the learning preprocessing unit 230 acquires the learning data candidate D21 from the learning sensor data acquisition unit 210 via the learning data storage unit 220, but this is only an example. The learning preprocessing unit 230 can acquire the learning data candidate D21 directly from the learning sensor data acquisition unit 210.

[0170] The reference structure learning unit 240 receives the learning data D23 output by the learning preprocessing unit 230 from the learning data storage unit 220 and learns the reference structure D2 based on the learning data D23.

[0171] In particular, the reference structure learning unit 240 calculates at least one statistic that indicates a relationship between two different parts of sensor data for a multitude of parts of sensor data contained in the learning data D23, and learns the reference structure D2 based on the calculated statistic.

[0172] The Reference Structure Learning Unit 240 uses a correlation or cross-correlation, or a waveform-based statistical index such as Granger causality, transfer entropy, convergent cross mapping (CCM), or dynamic time warping (DTW), as an index when calculating a statistic (hereinafter referred to as the "statistical index") that indicates a relationship between the sensor data. Additionally, the Reference Structure Learning Unit 240 can use a distribution-based statistical index such as Kullback-Leibler divergence (KL) or histogram intersection (HI) as a statistical index. The statistical index is categorized into undirected and directed types.The non-directional statistical index refers to a statistical index such as correlation, where the direction of the dependency relationship cannot be identified, and the directional statistical index refers to a statistical index such as Granger causality, where the direction of the dependency relationship can be identified.

[0173] In the first embodiment, for example, the reference structure learning unit 240 computes one or more types of statistics, including at least one directed statistical index that indicates a relationship between two different parts of sensor data for a plurality of parts of sensor data contained in the learning data D23, and learns the reference structure D2 based on the computed statistics.

[0174] The learning process in which the reference structure learning unit 240 learns the reference structure D2 is described using a concrete example and the drawings.

[0175] Fig. Figure 12 is a representation illustrating a concept of an example of learning processing in which the reference structure learning unit 240 learns the reference structure D2 in the first embodiment.

[0176] As in Fig. As shown in Figure 12, the training data D23 is a two-dimensional data frame in which one row represents the number t of time points and one column represents the number n of sensors 300 (represented as a sensor Xn (n = 1, 2, 3, 4, ..., n)). In the reference structure D2, all elements defined in the reference structure D2 are initialized with "0". It is assumed that the reference structure training unit uses 240 m types of statistical indices when learning the reference structure D2. <lerndatenauswahl>

[0177] The reference structure learning unit 240 selects sensor data collected by two different sensors Xn from the sensor data collected by the sensors Xn contained in the training data D23. For example, the reference structure learning unit 240 selects the i-th sensor data collected by the i-th sensor Xi and the j-th sensor data collected by the j-th sensor Xj. Hereafter, the i-th sensor data collected by the i-th sensor Xi will also be referred to simply as "i-th sensor data," and the j-th sensor data collected by the j-th sensor Xj will also be referred to simply as "j-th sensor data." <Berechnung der Statistik>

[0178] Next, the reference structure learning unit 240 computes the statistics a(1)ij, a(2)ij, ..., a(m)ij from the i-th sensor data to the j-th sensor data as the statistics from sensor Xi to sensor Xj, and computes statistics a(1)ij, a(2)ij, ..., a(m)ij from the i-th sensor data to the j-th sensor data as the statistics from sensor Xi to sensor Xj using m types of statistical indices. Then, the reference structure learning unit 240 acquires information (hereafter referred to as "statistical information between sensors") D24A regarding the statistics between sensor Xi and sensor Xj, including a statistic a(k)ij from sensor Xi to sensor Xj and a statistic a(k)ji from sensor Xj to sensor Xi.The reference structure learning unit 240 transforms the statistical information D24A between the sensors as needed, such that the dependency relationship between the parts of the sensor data increases with increasing absolute value | a(k)ij | of the statistic. For example, a p-value corresponding to the statistic a(k)ij of Granger causality takes on a value between 0 and 1, and the smaller the p-value, the less certain it is that there is no dependency relationship between the i-th and j-th sensor data. In this case, the reference structure learning unit 240 sets a statistic a(k)ij in the statistical information between the sensors as a (1-p-value), which is obtained by converting the p-value in such a way that the dependency relationship is stronger the larger the statistic is.On the other hand, in a case where the reference structure learning unit 240 calculates the statistics using correlation, which is the non-directional statistical index, for example, a correlation coefficient ρ, corresponding to the correlation statistic a(k)ij, takes a value between -1 and 1, indicating that the dependency relationship between the i-th sensor data and the j-th sensor data is greater the larger the absolute value of ρ. In this case, the reference structure learning unit 240 does not transform the statistical information D24A between the sensors.

[0179] It should be noted that the reference structure learning unit creates 240 pairs of two different parts of sensor data of all combinations among the sensor data contained in the learning data D23, calculates the statistics for all two different sensor data, and acquires the statistical information between the sensors D24A. <bezugsstrukturlernen>

[0180] The reference structure learning unit 240 learns the reference structure D2 using the statistics that correspond to all pairs of sensor data defined as one element in the statistical information D24A between the sensors. For example, the statistics |a(k)ij| from the i-th sensor Xi to the j-th sensor Xj, calculated using the k-th statistical index among the m types of statistical indices, are inserted into the k-th element in the first dimension, the i-th element in the second dimension, and the j-th element in the third dimension of the reference structure D2.

[0181] After learning the reference structure D2, as described above, the reference structure learning unit 240 causes the learned reference structure D2 to be stored in the data storage unit 20 of the anomaly factor calculation device 100.

[0182] For example, the reference structure learning unit 240 can store the reference structure D2 in the training data storage unit 220. In this case, in the anomaly factor calculation device 100, for example, the anomaly propagation path tracking unit 50 downloads the reference structure D2 to be used from the training data storage unit 220 to the data storage unit 20 each time the anomaly propagation sequence calculation processing is performed.

[0183] It should be noted that this assumes the reference structure learning unit 240 acquires the learning data D23 from the learning preprocessing unit 230 via the learning data storage unit 220, but this is only an example. The reference structure learning unit 240 can also acquire the learning data D23 directly from the learning preprocessing unit 230.

[0184] The description returns to the one in Fig. The configuration example of the learning device 200 shown in 11 is shown.

[0185] The learning data storage unit 220 stores various types of information about the learning performed by the learning device 200.

[0186] Specifically, the learning data storage unit 220 stores, for example, the learning data candidates D21 acquired by the learning sensor data acquisition unit 210 and the learning data D23 output by the learning preprocessing unit 230. The learning data storage unit 220 can store the reference structure D2 learned by the reference structure learning unit 240.

[0187] Furthermore, the learning data storage unit 220 is provided here in the learning device 200, but this is only an example, and the learning data storage unit 220 may be provided in a location that the learning device 200 can access outside of the learning device 200.

[0188] Furthermore, in the first embodiment, the learning device 200 includes the learning preprocessing unit 230, but this is only an example, and the learning device 200 need not necessarily include the learning preprocessing unit 230. In a case where the learning device 200 does not include the learning preprocessing unit 230, the reference structure learning unit 240, for example, sets all learning data candidates D21 acquired by the learning sensor data acquisition unit 210 as the learning data D23 actually used to learn the reference structure D2, and learns the reference structure D2 using the learning data D23 acquired by the learning sensor data acquisition unit 210.That is, in the learning device 200, the reference structure learning unit 240 defines a multitude of learning data candidates, acquired by the learning sensor data acquisition unit 210, as a multitude of parts of learning data, calculates at least one statistic between the multitude of parts of learning data based on the multitude of parts of learning data, and learns the estimated or calculated structure (reference structure D2) that indicates the dependency relationship between the plant components based on the calculated statistic.

[0189] The functionality of the anomaly factor calculation device 100 and the learning device 200 according to the first embodiment is described.

[0190] First, the functionality of the anomaly factor calculation device 100 according to the first embodiment is described.

[0191] Fig. Figure 13 is a flowchart describing the operation of the anomaly factor calculation device 100 according to the first embodiment.

[0192] The sensor data acquisition unit 10 acquires the sensor data D1 from the sensor 300 (step ST1).

[0193] The sensor data acquisition unit 10 initiates the storage of the acquired sensor data D1 in the data storage unit 20.

[0194] The anomaly detection unit 30 processes the sensor data D1, which is stored in the data storage unit 20 by the sensor data acquisition unit 10 in step ST1 (step ST2).

[0195] The anomaly detection unit 30 causes the anomaly detection sensor information D3 and the anomaly detection time information D4 to be stored in the data storage unit 20.

[0196] If the anomaly detection unit 30 detects an anomaly detection sensor in step ST2, the anomaly factor calculation device 100 proceeds to step ST3. If the anomaly detection unit 30 did not detect the anomaly detection sensor in step ST2, the anomaly factor calculation device 100 terminates the process shown in the flowchart of Fig. 13 processing shown.

[0197] For example, if the anomaly detection unit 30 detects the anomaly detection sensor in step ST2, the anomaly detection sequence calculation unit 40 is notified of the detection, and the operation of the anomaly factor calculation device 100 only needs to continue with step ST3. However, if the anomaly detection unit 30 did not detect the anomaly detection sensor in step ST2, it is sufficient for a control unit (not shown) of the anomaly factor calculation device 100 to be informed of the non-detection, and the control unit terminates the processing of the anomaly factor calculation device 100.

[0198] In step ST3, the anomaly detection sequence calculation unit 40 acquires the anomaly detection sensor information D3 and the anomaly detection time information D4, which were stored by the anomaly detection unit 30 in step ST2 in the data storage unit 20, and performs an anomaly detection sequence calculation to calculate the anomaly detection sequence in which the occurrence of an anomaly was detected in the anomaly detection sensor, more precisely, the occurrence of an anomaly was detected in the sensor data D1 that was collected by the anomaly detection sensor (step ST3).

[0199] The anomaly detection sequence calculation unit 40 causes the anomaly detection sequence calculation result D5 to be stored in the data storage unit 20.

[0200] The anomaly propagation path tracking unit 50 acquires the anomaly detection sensor information D3 and the reference structure D2, which are stored by the anomaly detection unit 30 from the data storage unit 20 in step ST3, and performs the anomaly propagation sequence calculation processing of the estimation or calculation of the anomaly propagation sequence based on the acquired anomaly detection sensor information D3 and the reference structure D2 (step ST4).

[0201] The anomaly propagation path tracking unit 50 outputs the anomaly propagation sequence calculation result D6 to the data storage unit 20.

[0202] The anomaly factor calculation unit 60 receives the anomaly detection sequence calculation result D5, which is output by the anomaly detection sequence calculation unit 40 in step ST3, and the anomaly propagation sequence calculation result D6, which is output by the anomaly propagation path tracking unit 50 in step ST4, from the data storage unit 20, and performs the anomaly factor calculation processing of calculating a factor of the anomaly based on the sequence of anomaly detection calculated by the anomaly detection sequence calculation unit 40 and the anomaly propagation sequence calculated by the anomaly propagation path tracking unit 50 (step ST5).

[0203] The anomaly factor calculation unit 60 generates the anomaly factor sequence calculation result D7 and causes the generated anomaly factor sequence calculation result D7 to be stored in the data storage unit 20.

[0204] The anomaly factor calculation result output unit 70 acquires from the data storage unit 20 the anomaly factor sequence calculation result D7, which is output by the anomaly factor calculation unit 60 in step ST5, the anomaly detection sequence calculation result D5, which is output by the anomaly detection sequence calculation unit 40 in step ST3, and the anomaly propagation sequence calculation result D6, which is output by the anomaly propagation path tracking unit 50 in step ST4, and outputs the information regarding the result of the calculation of the anomaly factor by the anomaly factor calculation unit 60 (step ST6).

[0205] In particular, the anomaly factor calculation result output unit 70 outputs the anomaly factor calculation result display information to the display device 400 for displaying the anomaly factor calculation result screen. The anomaly factor calculation result output unit 70 provides the operator with information about the result of the anomaly factor calculation.

[0206] It should be noted that in a case where the anomaly factor calculation device 100 does not contain the anomaly factor calculation result output unit 70, the anomaly factor calculation device 100 will process step ST6 in the operation of the anomaly factor calculation device 100, which is shown in the flowchart of Fig. The step shown in Figure 13 can be omitted. For example, the processing in step ST6 is performed by a device outside the anomaly factor calculation device 100.

[0207] As described above, the anomaly factor calculation device 100 detects a plurality of anomaly detection sensors based on a plurality of time series sensor data acquired by the plurality of sensors 300 provided in the target plant, and calculates the anomaly detection sequence in which the occurrence of an anomaly was detected for the plurality of anomaly detection sensors based on the detection time at which the plurality of anomaly detection sensors are detected.The anomaly factor calculation device 100 calculates the anomaly propagation sequence in which the anomaly has propagated, based on the anomaly detection sensor information relating to the multiple anomaly detection sensors and the calculated structure (reference structure) that shows the dependency relationship between the multiple components of which the target plant consists, and calculates the factor of the anomaly based on the calculated anomaly detection sequence and the anomaly propagation sequence.

[0208] Since the anomaly factor calculation device 100 estimates or calculates the factor of the anomaly that has occurred in the target plant based on the sequence in which the anomaly occurred and the sequence in which the anomaly is propagating, it is possible to calculate the factor of the anomaly more appropriately using a variety of criteria.

[0209] Since the anomaly factor calculation device 100 calculates the anomaly propagation sequence based on the calculated structure (reference structure), the calculation of the anomaly factor can be performed at an early stage without having to wait until sensor data is collected that is sufficient to build the reference structure D2 at the time of diagnosis of the target plant.

[0210] This means that the anomaly factor calculation device 100 can determine the factor of anomalies that have occurred in the target plant, regardless of the complexity or size of the target plant.

[0211] In addition, the anomaly factor calculation device 100 also outputs a result of the calculation of the anomaly factor.

[0212] Therefore, the Anomaly Factor Calculation Device 100 improves the interpretability and explainability of the anomaly factor calculation result for the operator. The Anomaly Factor Calculation Device 100 can reduce unnecessary inspection work by the operator and decrease their workload. Furthermore, the Anomaly Factor Calculation Device 100 can estimate the anomaly factor using a quantitative index that is independent of human subjectivity and provide a rationale for the calculation. The operator can thus establish a test sequence for the system with less effort.

[0213] In addition, the anomaly factor calculation device 100 determines the anomaly detection sensor using a univariate anomaly detection method such as the Hotelling theory or Discord.

[0214] Therefore, the anomaly factor calculation device 100 can better detect an anomaly in which only a portion of sensor data D1 changes. For example, an anomaly in which only a portion of sensor data D1 changes is one detected in the sensor data D1 acquired by one sensor 300 alone and is not related to another sensor 300.

[0215] Furthermore, the anomaly factor calculation device 100 captures the anomaly detection sensor using a multivariate anomaly detection method such as Graphical Lasso.

[0216] Therefore, the anomaly factor calculation device 100 can better detect an anomaly in which the relationship between the multiple parts of sensor data D1 changes. An anomaly in which the ratio between the multiple parts of sensor data D1 changes is, for example, an anomaly that has occurred in the sensor data D1 acquired by sensor 300 located in the upstream device, and an anomaly that also occurs in the sensor data D1 acquired by sensor 300 located in the downstream device, when two system components, here devices, are in a control relationship. For example, if an anomaly occurs in a valve opening degree acquired by a valve that regulates a certain flow rate, an anomaly will also occur in a flow rate measured by a flow meter that measures the flow rate.

[0217] Next, the functionality of the learning device 200 according to the first embodiment will be described.

[0218] Fig. Figure 14 is a flowchart describing the operation of the learning device 200 according to the first embodiment.

[0219] The learning sensor data acquisition unit 210 acquires the learning data candidate D21, which is used to learn the reference structure D2 (step ST21). Specifically, the learning sensor data acquisition unit 210 acquires learning data candidates comprising a multitude of time-series sensor data acquired by a plurality of sensors located in the target plant during the time of normal operation of the target plant.

[0220] The learning sensor data acquisition unit 210 initiates the storage of the acquired learning data candidate in the learning data storage unit 220.

[0221] The learning preprocessing unit 230 acquires the learning data candidate D21, acquired by the learning sensor data acquisition unit 210 in step ST21, from the learning data storage unit 220, performs preprocessing of the learning data candidate D21 and acquires the learning data D23 (step ST22).

[0222] The learning preprocessing unit 230 initiates the storage of the acquired learning data D23 in the learning data storage unit 220.

[0223] The reference structure learning unit 240 acquires the learning data D23 output by the learning preprocessing unit 230 from the learning data storage unit 220 and learns the reference structure D2 based on the learning data D23 (step ST23).

[0224] After learning the reference structure D2, the reference structure learning unit 240 causes the learned reference structure D2 to be stored in the data storage unit 20 of the anomaly factor calculation device 100.

[0225] Fig. 15 is a flowchart describing details of the processing of step ST23 in Fig. 14.

[0226] The reference structure learning unit 240 selects sensor data collected by two different sensors 300 from the sensor data contained in the training data D23 based on the training data D23, which was generated by the training preprocessing unit 230 in step ST22. Fig. 14 were acquired.

[0227] That is, the reference structure learning unit 240 procures a pair of two different pieces of sensor data based on the training data D23 (step ST231).

[0228] It should be noted that the reference structure learning unit 240 defines all combinations of a multitude of parts of sensor data contained in the learning data D23 as pairs of sensor data.

[0229] The reference structure learning unit 240 extracts the sensor data pair generated in step ST231 and calculates at least one statistic between two different parts of sensor data (step ST232).

[0230] The reference structure learning unit 240 calculates statistics for all two different parts of sensor data and acquires the statistical information between the sensors D24A.

[0231] Then the reference structure learning unit 240 learns the reference structure D2 by using the statistics that correspond to all pairs of sensor data set as one element in the statistical information between sensors D24A (step ST233).

[0232] It should be noted that in a case where the learning device 200 does not contain the learning preprocessing unit 230, the learning device 200 will process step ST22 in the operation of the learning device 200, which is shown in the flowchart of Fig. 14 is shown, can be omitted.

[0233] As described above, the learning device 200 acquires the multitude of time-series sensor data fragments acquired as training data candidates by the majority of sensors 300 supplied to the target plant during normal operation of the target plant, and procures the multitude of training data fragments used for learning based on the multitude of training data candidates. Based on the acquired training data, the learning device 200 calculates at least one statistic from among the multitude of sensor data fragments contained in the training data and learns the estimated or calculated structure (reference structure D2) based on the calculated statistic.

[0234] For example, the completeness of the manually specified reference structure depends on a connection relationship between plant components, which is detected by a person, i.e., sensor 300. On the other hand, the learning device 200 can comprehensively extract the relevance between the sensor data D1 and, as a result, provide the reference structure D2, in which the oversight of the connection relationship of sensor 300 is suppressed. The learning device 200 can cause the anomaly factor calculation device 100 to track sensor 300, which is the source of the anomaly, more appropriately and improve the estimation or calculation accuracy of the anomaly factor by providing the anomaly factor calculation device 100 with the calculated structure (reference structure) while tracking sensor 300, which is the source of the anomaly.

[0235] Furthermore, even if a reference structure containing only qualitative information such as connection information is manually specified, the learning device 200 can quantitatively determine the extent of relevance or the direction of influence between parts of the sensor data D1 and generate and deliver a calculated structure (reference structure) that is able to more appropriately estimate the factor of the anomaly factors.

[0236] Furthermore, the learning device 200 calculates the statistics using a waveform-based statistical index such as correlation, Granger causality, or DTW.

[0237] Therefore, the learning device 200 can track the anomaly propagation based on the dependency relationship of a similarity in the waveform and provide the calculated structure (reference structure D2) that can more appropriately calculate the factor of the anomalies.

[0238] Furthermore, the learning device 200 calculates statistics using a distribution-based statistical index such as KL divergence or Hl.

[0239] Therefore, the learning device 200 can track the anomaly propagation based on the dependency relationship of a similarity in the distribution and provide the calculated structure (reference structure D2) with which the anomaly factor can be calculated more appropriately.

[0240] Furthermore, the learning device 200 calculates the statistics using the waveform-based statistical index and the distribution-based statistical index.

[0241] Therefore, the learning device 200 can track the anomaly propagation based on the dependency relationship of the similarity of the waveform or distribution and provide the calculated structure (reference structure D2) with which the anomaly factor can be calculated more appropriately.

[0242] Fig. 16A and Fig. Figure 16B shows block diagrams illustrating an example of a hardware configuration of the anomaly factor calculation device 100 according to the first embodiment.

[0243] In the first embodiment, the functions of the sensor data acquisition unit 10, the anomaly detection unit 30, the anomaly detection sequence calculation unit 40, the anomaly propagation path tracking unit 50, the anomaly factor calculation unit 60, the anomaly factor calculation result output unit 70, and the control unit (not shown) are performed by the processing circuit 1601. That is, the anomaly factor calculation device 100 includes the processing circuit 1601 for calculating a factor of the anomaly that occurred in the target plant, using the calculated structure (reference structure D2) that specifies the dependency relationship between the majority of plant components that constitute the target plant.

[0244] The 1601 processing circuit may be dedicated hardware, as in Fig. 16A, or to represent a processor 1604 executing a program stored in memory, as in Fig. 16B shown.

[0245] In a case where the processing circuit 1601 is dedicated hardware, the processing circuit 1601 corresponds, for example, to a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination thereof.

[0246] If the processing circuit is the processor 1604, then the functions of the sensor data acquisition unit 10, the anomaly detection unit 30, the anomaly detection sequence calculation unit 40, the anomaly propagation path tracking unit 50, the anomaly factor calculation unit 60, the anomaly factor calculation result output unit 70, and the control unit (not shown) are implemented by software, firmware, or a combination of software and firmware. The software or firmware is described as a program and stored in a memory 1605.The processor 1604 reads and executes the program stored in memory 1605, thereby performing the functions of the sensor data acquisition unit 10, the anomaly detection unit 30, the anomaly propagation sequence calculation unit 40, the anomaly propagation path tracking unit 50, the anomaly factor calculation unit 60, the anomaly factor calculation result output unit 70, and the control unit (not shown). That is, the anomaly factor calculation device 100 contains memory 1605 for storing programs which, when executed by the processor 1604, perform the steps ST1 to ST6 described above. Fig. 13. It can also be said that the program stored in memory 1605 causes a computer to execute processing procedures or methods carried out by the sensor data acquisition unit 10, the anomaly detection unit 30, the anomaly detection sequence calculation unit 40, the anomaly propagation path tracking unit 50, the anomaly factor calculation unit 60, the anomaly factor calculation result output unit 70, and the control unit (not shown).Here, memory 1605 corresponds, for example, to non-volatile or volatile semiconductor memory such as RAM, read-only memory (ROM), flash memory, erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM) (registered trademark, omitted below), or a magnetic disk, a flexible disk, an optical disk, a compact disk, a mini-disk, a digital versatile disk (DVD), or similar.

[0247] Furthermore, the functions of the sensor data acquisition unit 10, the anomaly detection unit 30, the anomaly detection sequence calculation unit 40, the anomaly propagation path tracking unit 50, the anomaly factor calculation unit 60, the anomaly factor calculation result output unit 70 and the control unit, which is not shown, can be implemented partly by special hardware and partly by software or firmware.For example, the functions of the sensor data acquisition unit 10 and the anomaly factor calculation result output unit 70 can be implemented by the processing circuit 1601 as dedicated hardware, and the functions of the anomaly detection sequence calculation unit 40, the anomaly propagation path tracking unit 50, the anomaly factor calculation unit 60 and the control unit, which is not shown, can be implemented by the processor 1604, which reads and executes a program stored in memory 1605.

[0248] The data storage unit 20 includes an auxiliary memory (not shown).

[0249] Furthermore, the anomaly factor calculation device 100 includes an input interface device 1602 and an output interface device 1603, which perform wired or wireless communication with a device such as the sensor 300 or the display device 400.

[0250] An example of the hardware configuration of the learning device 200 according to the first embodiment is also included in the Fig. 16A and Fig. 16B shown.

[0251] In the first embodiment, the functions of the learning sensor data acquisition unit 210, the learning preprocessing unit 230, and the reference structure learning unit 240 are performed by the processing circuit 1601. That is, the learning device 200 contains the processing circuit 1601 for performing the control to learn the calculated structure (reference structure D2), which indicates the dependency relationship between the plurality of plant components comprising the target plant, based on the plurality of time-series sensor data acquired by the plurality of sensors 300 provided in the target plant during normal operation.

[0252] The 1601 processing circuit may be dedicated hardware, as in Fig. 16A, or to represent a processor 1604 executing a program stored in memory, as in Fig. 16B shown.

[0253] In a case where the processing circuit 1601 is dedicated hardware, the processing circuit 1601 corresponds, for example, to a single circuit, a composite circuit, a programmed processor, a parallel programmed processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or a combination thereof.

[0254] If the processing circuit is the processor 1604, the functions of the learning sensor data acquisition unit 210, the learning preprocessing unit 230, and the reference structure learning unit 240 are implemented by software, firmware, or a combination of software and firmware. The software or firmware is described as a program and stored in a memory 1605. The processor 1604 executes the functions of the learning sensor data acquisition unit 210, the learning preprocessing unit 230, and the reference structure learning unit 240 by reading and executing the program stored in memory 1605. That is, the learning device 200 contains memory 1605 for storing a program which, when executed by the processor 1604, performs steps ST21 to ST23 described above. Fig. 14. It can also be said that the program stored in memory 1605 causes a computer to execute processing procedures or methods performed by the learning sensor data acquisition unit 210, the learning preprocessing unit 230, and the reference structure learning unit 240. Here, memory 1605 corresponds to non-volatile or volatile semiconductor memory such as RAM, read-only memory (ROM), flash memory, erasable programmable read-only memory (EPROM), or electrically erasable programmable read-only memory (EEPROM), or a magnetic disk, flexible disk, optical disk, compact disk, mini-disk, digital versatile disc (DVD), or the like.

[0255] Furthermore, the functions of the learning sensor data acquisition unit 210, the learning preprocessing unit 230, and the reference structure learning unit 240 can be implemented partly by dedicated hardware and partly by software or firmware. For example, the functions of the learning sensor data acquisition unit 210 can be implemented by the processing circuit 1601 as dedicated hardware, and the functions of the learning preprocessing unit 230 and the reference structure learning unit 240 can be implemented by the processor 1604, which reads and executes programs stored in memory 1605.

[0256] The learning data storage unit 220 includes an auxiliary storage unit (not shown).

[0257] Furthermore, the learning device 200 includes an input interface device 1602 and an output interface device 1603, which perform wired or wireless communication with a device such as the anomaly factor calculation device 100.

[0258] Furthermore, for the sake of simplicity, the first embodiment described above assumes that one sensor 300 is present in a device, but this is only an example, and a plurality of sensors 300 can be present in a device. In the first embodiment described above, for example, the system component has a plurality of components contained in the device, and the sensor 300 can be provided in each component. In this case, the anomaly factor calculation device 100 provides the operator with information about the result of the anomaly factor calculation in a form that allows, for example, the identification of a sensor in a component that is the anomaly factor, or a sequence in which the operator must perform an inspection. <modifikation>

[0259] In the first embodiment described above, the anomaly factor calculation device 100 and the learning device 200 each comprise the sensor data acquisition unit 10 and the learning sensor data acquisition unit 210, but this is only an example. In the first embodiment described above, the anomaly factor calculation device 100 and the learning device 200 each comprise the data storage unit 20 and the learning data storage unit 220, but this is only an example.

[0260] For example, the anomaly factor calculation device 100 and the learning device 200 can contain a sensor data acquisition unit and a data storage unit that are common and can be set up to access each other.

[0261] Fig. Figure 17 is a block diagram illustrating a configuration example of a precise diagnostic system 1000 in which the anomaly factor calculation device 100 and the learning device 200 include a sensor data acquisition unit 310 and a data storage unit 320, which are common in the first embodiment.

[0262] Furthermore, the anomaly factor calculation device comprises 100, even if it is in Fig. Figure 17, which is not shown for the sake of simplicity, includes the anomaly detection unit 30, the anomaly detection sequence calculation unit 40, the anomaly propagation path tracking unit 50, the anomaly factor calculation unit 60, the anomaly factor calculation result output unit 70, and the control unit, in addition to the data storage unit 320. Furthermore, the anomaly factor calculation device 100 does not necessarily include the anomaly factor calculation result output unit 70. Moreover, the learning device 200 includes, besides the sensor data acquisition unit 310, the learning preprocessing unit 230 and the reference structure learning unit 240, although these are not shown for the sake of simplicity in Figure 17. Fig. 17 are not shown.

[0263] In the Fig. The configuration example shown in Figure 17 of the precise diagnostic system 1000 is merely an example, although the anomaly factor calculation device 100 contains the data storage unit 320 and the learning device 200 contains the sensor data acquisition unit 310, and in the precise diagnostic system 1000 the learning device 200 can contain the data storage unit 320 and the anomaly factor calculation device 100 can contain the sensor data acquisition unit 310.

[0264] Furthermore, the precise diagnostic system 1000 can include the sensor data acquisition unit 310 and the data storage unit 320, either from the anomaly factor calculation device 100 or the learning device 200. <modifikation>

[0265] In the first embodiment described above, the anomaly factor calculation device 100 is set up to calculate an anomaly factor based on a reference structure D2, but this is only an example.

[0266] For example, the anomaly factor calculation device 100 can be configured to calculate the anomaly factor based on a structure corresponding to the operating state of the target plant. In this case, the learning device 200 learns the reference structure for each operating state of the target plant.

[0267] Fig. Figure 18 is a block diagram illustrating a configuration example of the precise diagnostic system 1000, in which, in the first embodiment, the learning device 200 learns the reference structure for each operating state of the target plant and the anomaly factor calculation device 100 calculates the anomaly factor on the basis of the reference structure corresponding to the operating state of the target plant learned by the learning device 200.

[0268] Furthermore, the anomaly factor calculation device comprises 100, which are in Fig. Figure 18, which is not shown for the sake of simplicity, comprises the sensor data acquisition unit 10, the anomaly detection unit 30, the anomaly detection sequence calculation unit 40, the anomaly factor calculation unit 60, the anomaly factor calculation result output unit 70, and the control unit, in addition to the data storage unit 20 and the anomaly propagation path tracking unit 50. Furthermore, the anomaly factor calculation device 100 does not necessarily include the anomaly factor calculation result output unit 70. Additionally, the learning device 200 comprises the learning sensor data acquisition unit 210 and the learning preprocessing unit 230, in addition to the learning data storage unit 220 and the reference structure learning unit 240, although this is not explicitly stated in Figure 18. Fig. 18 is not shown for the sake of simplicity.

[0269] As in Fig. As shown in Figure 18, the reference structure learning unit 240 in the learning device 200 acquires the facility operating state information D31, which specifies the operating state of the target facility according to the learning data candidate D21. The operator or the like uses the input device, such as a mouse or a keyboard, to input the facility operating state information D31, and the reference structure learning unit 240 acquires the facility operating state information D31 by receiving the input facility operating state information D31. For example, the reference structure learning unit 240 can obtain the facility operating state information D31 by detecting a control signal or similar in the target facility and estimating the operating state from the detected control signal or similar.

[0270] Then, after learning the reference structure D2 based on the learning data D23 output by the learning preprocessing unit 230, the reference structure learning unit 240 assigns the acquired facility operating state information D31 to the learned reference structure D2 to obtain the reference structure D32. The reference structure learning unit 240 causes the reference structure D32, to which the facility operating state information D31 has been assigned, to be stored in the data storage unit 20 of the anomaly factor calculation device 100. The reference structure learning unit 240 can also cause the reference structure D32 to be stored in the learning data storage unit 220.

[0271] The learning device 200 performs the learning described above depending on different operating states of the target system and learns the reference structure D32, which corresponds to the different operating states.

[0272] In this case, the reference structure learning unit 240 acquires, with reference to the flowchart of Fig. As described in section 14, the learning device 200 retrieves the setup operating state information D31 before processing step ST23 and performs the processing of generating and storing the reference structure D32 in step ST23. The learning device 200 repeats the process as shown in the flowchart of Fig. 14 shown, depending on the operating state of the target plant.

[0273] In the anomaly factor calculation device 100, the anomaly propagation path tracking unit 50 acquires the facility operating state information D31. Then, when the anomaly propagation sequence calculation processing is performed, the anomaly propagation path tracking unit 50 uses the acquired anomaly detection sensor information D3, the facility operating state information D31, and the reference structure D32, which is stored in the data storage unit 20 by the learning device 200, to calculate the anomaly propagation sequence. Specifically, the anomaly propagation path tracking unit 50 selects the reference structure D32 that corresponds to the operating state of the target facility and calculates the anomaly propagation sequence using the selected reference structure D32.

[0274] For example, in the learning device 200, the reference structure learning unit 240 can cause the reference structure D32 to be stored in the learning data storage unit 220, and in the anomaly factor calculation device 100, the anomaly propagation path tracking unit 50 can download the reference structure D32 to be used from the learning data storage unit 220 to the data storage unit 20 each time the anomaly propagation sequence calculation is performed.

[0275] In this case, the anomaly propagation path tracking unit 50 determines the processes of the anomaly factor calculation device 100, which are based on the flowchart of Fig. 13 are described, in step ST4 the anomaly propagation sequence is based on the anomaly detection sensor information D3, the facility operating state information D31 and the reference structure D32, which are stored by the learning device 200 in the data storage unit 20.

[0276] As described above, the anomaly propagation path tracking unit 50 can be set up in the anomaly factor calculation device 100, which calculates the anomaly propagation sequence based on the anomaly detection sensor information D3, the facility operating state information D31, which specifies the operating state of the target plant, and the calculated structure (reference structure D32), which specifies the dependency relationship between a plurality of target components that make up the target plant, depending on the operating state of the target plant.

[0277] With such a configuration, the anomaly factor calculation device 100 can cope with a change in the dependency relationship between the sensors 300 due to the change in the operating state in the target plant and accurately determine the anomaly factor based on the reference structure D32 with improved reliability.

[0278] If the reference structure learning unit 240 calculates at least one statistic from the multitude of sensor data based on the training data and the calculated structure (reference structure D2) learns based on the calculated statistic, the learning device 200 generates the reference structure D32 in which the facility operating state information D31 is added to the reference structure D2, so that the reliability of the reference structure supplied to the anomaly factor calculation device 100 is improved and the reference structure D32, which is able to accurately calculate the factor of the anomaly, can be supplied to the anomaly factor calculation device 100. <modifikation>

[0279] In the first embodiment described above, the anomaly factor calculation device 100 comprises the data storage unit 20, but this is only an example.

[0280] For example, a single or multiple network storage devices (not shown) arranged in a communication network can store various data, and in the anomaly factor calculation device 100, the anomaly detection unit 30, the anomaly detection sequence calculation unit 40, the anomaly propagation path tracking unit 50, the anomaly factor calculation unit 60, and the anomaly factor calculation result output unit 70 can access the network storage device. <modifikation>

[0281] In the first embodiment described above, the anomaly detection unit 30 in the anomaly factor calculation device 100 performs the anomaly detection processing using the known univariate anomaly detection method on the sensor data D1 in order to detect the anomaly detection sensor among the sensors 300, but this is only an example.

[0282] For example, the anomaly detection unit 30 can detect the anomaly detection sensor using a known multivariate anomaly detection method. A known multivariate anomaly detection method is, for example, Graphical Lasso. For example, the anomaly detection unit 30 can perform anomaly detection processing using both univariate and multivariate anomaly detection methods.

[0283] As described above, the anomaly factor calculation device 100 detects the anomaly detection sensor using the univariate anomaly detection method, the multivariate anomaly detection method, or both methods, so that the anomaly detection sensor, in which the occurrence of various types of anomalies has been detected, can be adequately detected even if the way in which the anomaly appears in the sensor data differs depending on the type of anomaly occurring in the target plant, and the calculation of the anomaly factor can be performed accurately. <modifikation>

[0284] In the first embodiment described above, the anomaly factor calculation device 100 can include a reference structure correction unit 330, which corrects the reference structure D2 stored in the data storage unit 20.

[0285] Fig. Figure 19 is a block diagram showing a configuration example of the anomaly factor calculation device 100 including the reference structure correction unit 330 in the first embodiment.

[0286] Furthermore, the anomaly factor calculation device comprises 100, which are in Fig. Figure 19, omitted for simplicity, includes the sensor data acquisition unit 10, the anomaly detection unit 30, the anomaly detection sequence calculation unit 40, the anomaly factor calculation unit 60, the anomaly factor calculation result output unit 70, and the control unit, in addition to the reference structure correction unit 330, the data storage unit 20, and the anomaly propagation path tracking unit 50. Furthermore, the anomaly factor calculation device 100 does not necessarily include the anomaly factor calculation result output unit 70. Additionally, the anomaly factor calculation device 100 is connected to the learning device 200, the sensor 300, and the display device 400, although this is omitted for simplicity. Fig. 19 is not shown.

[0287] The reference structure correction unit 330 acquires information (hereinafter referred to as "dependent pair information") D33 about a sensor pair with a dependency relationship and information (hereinafter referred to as "non-dependent pair information") D34 about a sensor pair without a dependency relationship. The dependent pair information D33 and the non-dependent pair information D34 can, for example, be generated manually by the operator based on their expertise, or they can be generated by the reference structure correction unit 330 from information specifying a physical connection relationship, such as design information of the facility.For example, the information about dependent pairs D33 and the information about non-dependent pairs D34 can be generated by the reference structure correction unit 330 through a combination of manual generation based on the operator's know-how and generation based on information that indicates a physical connection relationship, such as design information of the facility.

[0288] The reference structure correction unit 330 obtains the reference structure D2 from the data storage unit 20, corrects the dependency relationship between the sensor data for reference structure D2 based on the information about dependent pairs D33 and the information about non-dependent pair D34, and causes the corrected reference structure D35 to be stored in the data storage unit 20. Furthermore, after the correction, reference structure D35 has the same data structure as reference structure D2.

[0289] When the corrected reference structure D35 is stored in the data storage unit 20, the anomaly propagation path tracking unit 50 performs the anomaly propagation sequence calculation processing using the corrected reference structure D35.

[0290] The details of the processing of the correction of the reference structure D2 by the reference structure correction unit 330 are described using a specific example.

[0291] Fig. Figure 20 is a representation to describe a concept of a processing example in which the reference structure correction unit 330 corrects the reference structure D2 in the anomaly factor calculation device 100, which contains the reference structure correction unit 330 in the first embodiment.

[0292] In this example, the number of sensors 300 is set to three, and sensor 300 is represented by a sensor Xn (n = 1, 2, 3). Furthermore, the dependency pair information D33 defines that there is a dependency relationship between the i-th sensor Xi and the j-th sensor Xj among the sensors Xn, and the non-dependent pair information D34 defines that there is no dependency relationship between the i-th sensor Xi and the j-th sensor Xj among the sensors Xn. i and j are 1, 2, and 3, respectively.

[0293] For example, it is assumed that there is a dependency relationship between sensors X1, X2 and X3 between sensor data from sensor X1 to sensor X2 and between sensor X3 and sensor X2, while there is no dependency relationship between sensor data from sensor X2 to sensor X1.

[0294] The reference structure D2 is assumed to be a three-dimensional field where the first dimension is a statistical index, the second dimension is the number of sensors Xn, and the third dimension is the number of sensors Xn. There are two types of statistical indices.

[0295] The reference structure correction unit 330 corrects the statistic a(k)ij of reference structure D2 based on the information about dependent pairs D33 and the information about non-dependent pairs D34 to a statistic a'(k)ij. Here, the statistic a(k)ij and the statistic a'(k)ij are real numbers. For example, the reference structure correction unit 330 corrects the statistic a(k)ij of reference structure D2, which corresponds to the sensor pair Xi and Xj contained in the information about dependent pairs D33, to a statistic a'(k)ij that is greater than an upper limit of the statistic, indicating that a dependency relationship exists for each statistical index.Furthermore, this is only one example, and for instance, the reference structure correction unit 330 can correct the statistic a(k)ij of the reference structure D2, which corresponds to the sensor pair Xi and Xj contained in the information about dependent pairs D33, to a statistic a'(k)ij that is greater than a threshold provided for the selection of the dependency relationship.

[0296] In Fig. 20. Based on the information about dependent pairs D33, the reference structure correction unit 330 corrects the statistics a(1)12 and a(2)12, which correspond to the sensor data from sensor X1 to sensor X2 and have a dependency relationship with the statistics a'(1)12 and a'(2)12, which are upper limits of the statistical index. Furthermore, the reference structure correction unit 330 corrects the statistics a(1)32 and a(2)32, which correspond to the sensor data from sensor X3 to sensor X2 and have a dependency relationship with the statistics a'(1)32 and a'(2)32, which are upper limits of the statistical index, based on the information about dependent pairs D33.

[0297] Furthermore, the reference structure correction unit 330 corrects the statistic a(k)ij of the reference structure D2, which corresponds to the pair of sensors Xi and Xj contained in the information D34 about the non-dependent pair, to the statistic a'(k)ij, which indicates that there is no dependency relationship for each statistical index.

[0298] In Fig. 20 The reference structure correction unit 330 corrects the statistics a(1)21 and a(2)21, which correspond to the sensor data from sensor X2 to sensor X1 and have no dependency relationship, to statistics a'(1)21 and a'(2)21, which indicate that there is no dependency relationship based on the information about non-dependent pairs D34.

[0299] As in Fig. As shown in Figure 19, the reference structure correction unit 330 corrects in a case where the anomaly factor calculation device 100 contains the reference structure correction unit 330, with reference to the flowchart of Fig. The operation of the anomaly factor calculation device 100, as described in section 13, determines the dependency relationship between the sensor data for the reference structure D2 based on the information about dependent pairs D33 and the information about non-dependent pairs D34, and causes the corrected reference structure D35 to be stored in the data storage unit 20 until the processing of step ST4 is performed. In step ST4, the anomaly propagation path tracking unit 50 determines an anomaly propagation sequence based on the anomaly detection sensor information D3, the facility operating state information D31, and the reference structure D35 corrected by the reference structure correction unit 330.

[0300] As described above, the anomaly factor calculation device 100 contains the reference structure correction unit 330, which corrects the dependency relationship between the sensor data for the calculated structure D2 on the basis of the information about dependent pairs D33, which refer to the pair of sensors 300 with a dependency relationship between the sensors 300, and the information about non-dependent pairs D34, which refer to the pair of sensors 300 without a dependency relationship, so that it is possible to improve the reliability of the calculated structure D2 and to calculate the anomaly factor accurately. <modifikation>

[0301] In the first embodiment described above, the anomaly factor calculation device 100 can include the relationship change calculation unit 340, which calculates a change in a relationship between the parts of sensor data by comparing the learned reference structure D2 with the reference structure D36 at the time an anomaly occurs. A change in a relationship between the sensor data presupposes, for example, a breakdown of the relationship between the sensor data. The relationship change calculation unit 340 determines a point at which the relationship between the parts of sensor data breaks down significantly in the element of the reference structure D2 as a point at which the relationship between the sensor data has changed.

[0302] Fig. Figure 21 is a block diagram showing a configuration example of the anomaly factor calculation device 100 including the relationship change calculation unit 340 in the first embodiment.

[0303] Furthermore, the anomaly factor calculation device comprises 100, even if it is in Fig. 21, which are not shown for the sake of simplicity, include the sensor data acquisition unit 10, the anomaly detection unit 30, the anomaly detection sequence calculation unit 40, the anomaly propagation path tracking unit 50, and the control unit, in addition to the relationship change calculation unit 340, the data storage unit 20, the anomaly factor calculation unit 60, and the anomaly factor calculation result output unit 70. Furthermore, the anomaly factor calculation device 100 does not necessarily have to include the anomaly factor calculation result output unit 70. Moreover, the learning device 200 includes the learning sensor data acquisition unit 210, in addition to the learning data storage unit 220, the learning preprocessing unit 230, and the reference structure learning unit 240, even if this is not shown in Figure 21. Fig. 21 is not shown for the sake of simplicity.

[0304] The relationship change calculation unit 340 acquires the reference structure D2, the reference structure D36 at the time of an anomaly, and the anomaly detection sensor information D3 from the data storage unit 20. Then, the relationship change calculation unit 340 compares the reference structure D2 and the reference structure D36 based on the reference structure D2, the reference structure D36, and the anomaly detection sensor information D3, and performs a relationship change order calculation to calculate an order (hereinafter referred to as the "relationship change order") of changes in the relationship between the parts of sensor data.

[0305] The reference structure D36 at the time of the anomaly is acquired through the following processing. In the learning device 200, the learning sensor data acquisition unit 210 acquires the sensor data D1 at the time of the anomaly's occurrence (including the period during which the anomaly is acquired) from the data storage unit 20 of the anomaly factor calculation device 100 and causes the sensor data D1 to be stored in the learning data storage unit 220. The learning preprocessing unit 230 acquires the sensor data D1 at the time of the anomaly's occurrence from the learning data storage unit 220, performs preprocessing on the acquired sensor data D1 at the time of the anomaly's occurrence, and outputs the sensor data D38 at the time of the anomaly's occurrence, after preprocessing, to the reference structure learning unit 240.Furthermore, the learning preprocessing unit 230 can output the sensor data D38 at the time of the anomaly occurrence, after preprocessing, to the reference structure learning unit 240 via the learning data storage unit 220. The reference structure learning unit 240 learns the reference structure D36 at the time of the anomaly based on the sensor data D38 at the time of the anomaly occurrence, after preprocessing, output by the learning preprocessing unit 230. The reference structure learning unit 240 only needs to learn the reference structure D36 in a similar way to the reference structure D2. The reference structure learning unit 240 learns the reference structure D36 at the time of the anomaly for the sensor data D1 at the time of the anomaly occurrence.

[0306] The reference structure learning unit 240 causes the learned reference structure D36 to be stored in the data storage unit 20 of the anomaly factor calculation device 100 at the time of the anomaly.

[0307] Details of the processing of the relationship change calculation unit 340 are described.

[0308] First, the relationship change calculation unit 340 acquires the reference structure D2 and the reference structure D36 at the time of the anomaly from the data storage unit 20, and from the reference structure D2 and the reference structure D36 at the time of the anomaly, it acquires a reference structure (hereinafter referred to as the "relationship change calculation reference structure") and a reference structure at the time of the anomaly (hereinafter referred to as the "relationship change calculation anomaly time reference structure"), which are reference structures corresponding to a type of non-directed statistical index that was selected to calculate a relationship change.Here, the relationship change calculation reference structure and the relationship change calculation anomaly time reference structure are two-dimensional matrices A(k) and A'(k), obtained by extracting only a portion corresponding to the k-th statistical index from each of the reference structures D2 and D36 at the time of the anomaly. Relationship change calculation unit 340 selects a statistical index meaningful for comparing statistics as the statistical index to be extracted. For example, relationship change calculation unit 340 determines that the statistical index calculated based on the p-value of the hypothesis tests is not a statistical index meaningful for comparing statistics and selects other statistical indices as statistical indices meaningful for comparing statistics. Relationship change calculation unit 340 can detect a relationship collapse, i.e.,a change in relationship, by comparing the statistics corresponding to the correlation, i.e., the magnitude of the correlation coefficient.

[0309] The relationship change calculation unit 340 then calculates a change amount d(kij) based on the captured element a(k)ij of the relationship change calculation reference structure and the element a'(k)ij of the relationship change calculation anomaly time reference structure. The relationship change calculation unit 340 then generates a reference structure change amount, which is information that contains the calculated change amount d(k)ij as one element and in which the change amount d(k)ij is displayed by a matrix.

[0310] Here, both the elements a(k)ij and a'(k)ij, as well as the change amount d(k)ij, are real numbers. The relationship change calculation unit 340 can calculate the change amount d(k)ij, which is an element of the reference structure change amount, for example, using an absolute value of a difference between an absolute value of the element a(k)ij of the relationship change calculation reference structure and an absolute value of the element a'(k)ij of the relationship change calculation abnormal time reference structure, or it can calculate the change amount d(k)ij using an absolute value of a difference between the element a(k)ij of the relationship change calculation reference structure and the element a'(k)ij of the relationship change calculation abnormal time reference structure.

[0311] Next, the relationship change calculation unit 340 acquires the anomaly detection sensor information D3 from the data storage unit 20 and calculates a relationship change degree for each sensor 300, more precisely for each anomaly detection sensor, based on the change amount d(k)ij, which is an element of the calculated relationship structure change amount, and the anomaly detection sensor information D3.

[0312] Here, both the change amount d(k)ij and the relationship change degree are real numbers. For example, the relationship change calculation unit 340 calculates the relationship change degree corresponding to the anomaly detection sensor Xi, which is the i-th sensor 300 among n sensors 300, by using an average of the elements that do not correspond to the i-th column and correspond to sensor 300 (i.e., the anomaly detection sensor) contained in the anomaly detection sensor information D3 in the i-th row of the reference structure change amount. Furthermore, this is only an example, and the relationship change calculation unit 340 can, for instance, calculate the relationship change degree corresponding to the anomaly detection sensor Xi, which is the i-th sensor 300 among the n sensors 300, by using an average of elements other than the i-th column in the i-th row of the reference structure change amount.

[0313] The Relationship Change Calculation Unit 340 then assigns the relationship change order corresponding to the anomaly detection sensor based on the calculated relationship change degree corresponding to the anomaly detection sensor. Both the relationship change degree and the relationship change order are real numbers. The Relationship Change Calculation Unit 340 assigns the relationship change order such that, for example, the corresponding relationship change order is in ascending order, starting with the anomaly detection sensor with the highest relationship change degree. Furthermore, if the relationship change degrees corresponding to multiple anomaly detection sensors are equal, the Relationship Change Calculation Unit 340 assigns the same relationship change order to the majority of the anomaly detection sensors.

[0314] After the relationship change sequence is assigned, the relationship change calculation unit 340 generates a relationship change sequence calculation result D37. The relationship change sequence calculation result is a piece of information in which the information specifying the anomaly detection sensor, the relationship change degree, and the relationship change sequence are combined.

[0315] The relationship change calculation unit 340 causes the relationship change sequence calculation result D37 to be stored in the data storage unit 20.

[0316] The relationship change sequence calculation processing by the relationship change calculation unit 340, as described above, is described with specific examples with reference to the drawings.

[0317] Fig. Figure 22 is a representation describing a concept of an example of a relationship change sequence calculation processing carried out by the relationship change calculation unit 340 on the basis of a learned reference structure D2 and a reference structure D36 at the time of an anomaly in a case where the anomaly factor calculation device 100 according to the first embodiment contains the relationship change calculation unit 340.

[0318] As an example, the number of sensors 300 is set to four, and sensor 300 is represented by sensor Xn (n = 1, 2, 3, 4). Furthermore, as an example, it is assumed that sensors X1, X2, and X3 are anomaly detection sensors.

[0319] First, the relationship change calculation unit 340 acquires the reference structure D2 and the reference structure D36 at the time of an anomaly from the data storage unit 20, and acquires a relationship change calculation reference structure D2A and a relationship change calculation anomaly time reference structure D36A from the reference structures D2 and D36 at the time of the anomaly. It should be noted that in Fig. 22 the reference structure D2 and the reference structure D36 are not shown at the time of the anomaly.

[0320] In Fig. Figure 22 shows the relationship change calculation reference structure D2A represented as a two-dimensional matrix obtained by extracting only a portion corresponding to the k-th statistical index from the reference structure D2. The relationship change calculation reference structure D2A is represented by a two-dimensional matrix where the first dimension is the set of four sensors Xn and the second dimension is the set of four sensors Xn. The statistical index is an undirected statistical index, and the type of statistical index is a correlation. As further explained in Fig. As shown in Figure 22, the relationship change calculation abnormal time reference structure D36A and the reference structure change amount D39 have the same data structure as the relationship change calculation reference structure D2A.

[0321] In Fig. As an example, the change amount d(k)ij, which is an element of the reference structure change amount D39, is an absolute value with respect to a difference between the absolute value of the element a(k)ij of the relationship change calculation reference structure D2A and the absolute value of the element a'(k)ij of the relationship change calculation abnormal time reference structure D36A. For example, the relationship change calculation unit 340 calculates the change amount between sensor X1 and sensor X2 as | |a(k)12| - |a'(k)12|= d(k)12 based on the element a(k)12 in the first row and second column of the relationship change calculation reference structure D2A and the element a'(k)12 in the first row and second column of the relationship change calculation abnormal time reference structure D36A. Here, | · | for an absolute value.

[0322] Furthermore, in Fig. 22 As an example, suppose that the relationship change calculation unit 340 calculates the relationship change degree dn corresponding to the anomaly detection sensor Xn by using the average of the elements that differ from the nth column and correspond to sensor 300 (i.e., the anomaly detection sensor) contained in the anomaly detection sensor information D3 in the nth row of the reference structure change amount D39. For example, relationship change calculation unit 340 sets the relationship change degree d1 corresponding to the anomaly detection sensor X1 as an average of the elements d(k)nn (with n = 1, 2, 3) corresponding to the anomaly detection sensors X1, X2, and X3 that differ from the first column in the first row of the reference structure change amount D39.Specifically, the relationship change calculation unit 340 calculates the relationship change degree d1 of the anomaly detection sensor X1 as the average of the change amount d(k)12, which is an element in the first row and second column, and the change amount d(k)13, which is an element in the first row and third column of the reference structure change amount D39, corresponding to the anomaly detection sensors X2 and X3. Similarly, the relationship change calculation unit 340 calculates the relationship change degrees d2 and d3, corresponding to sensors X2 and X3, respectively. Based on the calculated relationship change degrees d1, d2, and d3, the relationship change calculation unit 340 assigns the corresponding relationship change sequences o1, o2, and o3 to sensors X1, X2, and X3, respectively.

[0323] The relationship change calculation unit 340 then causes the relationship change sequence calculation result D37 to be stored in the data storage unit 20.

[0324] The anomaly factor calculation unit 60 receives the anomaly detection sequence calculation result D5, the anomaly propagation sequence calculation result D6 and the relationship change sequence calculation result D37 from the data storage unit 20 and performs the anomaly factor calculation processing taking into account the relationship change sequence based on the anomaly detection sequence calculation result D5, the anomaly propagation sequence calculation result D6 and the relationship change sequence calculation result D37.In particular, performing the anomaly factor calculation processing, taking into account the anomaly propagation sequence based on the anomaly detection sequence calculation result D5, the anomaly propagation sequence calculation result D6, and the relationship change sequence calculation result D37, means that the anomaly factor calculation unit 60 calculates a corresponding anomaly factor value from the anomaly detection sequence contained in the anomaly detection sequence calculation result D5, the anomaly propagation sequence contained in the anomaly propagation sequence calculation result D6, and the relationship change sequence contained in the relationship change sequence calculation result D37, and determines the anomaly factor sequence based on the calculated anomaly factor value.

[0325] Furthermore, the anomaly factor calculation unit 60 only needs to calculate the anomaly factor value from the anomaly detection sequence, the anomaly propagation sequence, and the relationship change sequence using a procedure similar to the procedure for calculating the anomaly factor value from the anomaly detection sequence and the anomaly propagation sequence.

[0326] The anomaly factor calculation unit 60 causes a result of the anomaly factor sequence calculation result D40 to be stored in the data storage unit 20, taking into account the relationship change sequence.

[0327] The anomaly factor calculation result output unit 70 receives the anomaly factor sequence calculation result D40, the anomaly detection sequence calculation result D5, the anomaly propagation sequence calculation result D6 and the relationship change sequence calculation result D37 from the data storage unit 20 and outputs information about the calculation result of the anomaly factor by the anomaly factor calculation unit 60.In particular, based on the anomaly factor calculation result output unit 70, the anomaly factor calculation result D40, the anomaly detection sequence calculation result D5, the anomaly propagation sequence calculation result D6, and the relationship change sequence calculation result D37, the anomaly factor calculation result display information is sent to the display device 400 to display the anomaly factor calculation result screen, which provides information regarding the calculation result of the anomaly factor determined by the anomaly factor calculation unit 60.

[0328] Furthermore, the anomaly factor device calculation unit 340 determines the relationship change calculation unit 340 in a case where the anomaly factor calculation device 100, as in Fig. 21 shows, in operation the anomaly factor calculation device 100, which, with reference to the flowchart of Fig. As described in section 13, a change in a relationship between the parts of sensor data occurs until the processing of step ST5 is carried out, and causes the relationship change sequence calculation result D37 to be stored in the data storage unit 20.

[0329] In step ST5, the anomaly factor calculation unit 60 acquires the anomaly detection sequence calculation result D5, the anomaly propagation sequence calculation result D6, and the relationship change sequence calculation result D37 from the data storage unit 20 and performs the anomaly factor calculation processing, taking into account the relationship change sequence, based on the anomaly detection sequence calculation result D5, the anomaly propagation sequence calculation result D6, and the relationship change sequence calculation result D37.

[0330] As described above, the anomaly factor calculation device 100 comprises the relationship change calculation unit 340, which compares the reference structure D2 (calculated structure) with the reference structure D36 at the time of the anomaly occurrence and calculates a change in a relationship between the parts of sensor data based on the relationship between reference structure D2 (calculated structure), the relationship between reference structure D36 at the time of the anomaly occurrence, and the anomaly detection sensor information D3. The anomaly factor calculation unit 60 is configured to calculate the anomaly factor taking into account the change in the relationship between the parts of sensor data, which is calculated by the relationship change calculation unit 340 based on the anomaly detection sequence calculated by the anomaly detection sequence calculation unit 40 and the anomaly propagation sequence.to determine the anomaly propagation path tracking unit 50, so that the reliability of the anomaly factor sequence calculation result D7 is improved and the anomaly factor can be calculated accurately. Since the anomaly factor calculation device 100 contains the relationship change calculation unit 340, a criterion for calculating an anomaly factor is added to the anomaly factor calculation device 100. <modifikation>

[0331] In the first embodiment described above, "calculating the anomaly factor" by the anomaly factor calculation device 100 means that the anomaly factor value, which indicates the degree of probability of the anomaly's source, and the anomaly factor sequence are calculated based on the anomaly factor value in units of sensors 300, and the information regarding the anomaly factor value and the anomaly factor sequence is generated. Furthermore, "calculating the anomaly factor" by the anomaly factor calculation device 100 can include calculating an anomaly factor value and an anomaly factor sequence based on the anomaly factor value in units of the device.

[0332] Fig. Figure 23 is a block diagram illustrating a configuration example of the anomaly factor calculation device 100, which includes an anomaly factor device calculation unit 350 and has a configuration for calculating an anomaly factor in units of the device in the first embodiment.

[0333] Furthermore, the anomaly factor calculation device comprises 100, although it is in Fig. Figure 23, which is not shown for the sake of simplicity, comprises the sensor data acquisition unit 10, the anomaly detection unit 30, the anomaly detection sequence calculation unit 40, the anomaly propagation path tracking unit 50, the anomaly factor calculation unit 60, and the control unit, in addition to the anomaly factor device calculation unit 350, the data storage unit 20, and the anomaly factor calculation result output unit 70. Furthermore, the anomaly factor calculation device 100 does not necessarily have to include the anomaly factor calculation result output unit 70. Moreover, the anomaly factor calculation device 100 is connected to the learning device 200, although this is not shown for the sake of simplicity. Fig. 23 is not shown.

[0334] The anomaly factor device calculation unit 350 acquires information D41 from sensors attached to a device. Furthermore, the anomaly factor device calculation unit 350 receives the anomaly detection sequence calculation result D5 and the anomaly propagation sequence calculation result D6 from the data storage unit 20. The anomaly factor device calculation unit 350 performs a device unit anomaly factor calculation process, in which an anomaly factor in device units is calculated based on the information D41 from sensors attached to a device, the anomaly detection sequence calculation result D5, and the anomaly propagation sequence calculation result D6.

[0335] The information D41 from sensors attached to a device is tabular data indicating in which device the sensor 300 is present. The operator or similar uses the input device, such as a mouse or keyboard, to input the information D41 from sensors attached to a device, and the anomaly factor device calculation unit 350 acquires the information D41 from sensors attached to a device by receiving the input information D41 from sensors attached to a device.

[0336] For example, in the information D41 of sensors attached to a device, information specifying a device is assigned to information specifying the sensor 300 present in the device.

[0337] The device unit anomaly factor calculation processing by the anomaly factor device calculation unit 350 is described in detail.

[0338] First, the anomaly factor device calculation unit 350 acquires the information D41 from sensors attached to a device, the anomaly detection sequence calculation result D5 and the anomaly propagation sequence calculation result D6.

[0339] The anomaly factor device calculation unit 350 converts the anomaly detection sequence calculation result D5 into an anomaly detection sequence calculation result (hereinafter referred to as "device anomaly detection sequence calculation result") D42 in units of the device based on the information D41 from sensors attached to a device and the anomaly detection sequence calculation result D5.

[0340] Specifically, the anomaly factor device calculation unit 350 compares information specifying sensor 300, assigned in the information D41 of sensors attached to a device, with information specifying sensor 300, which is included in the anomaly detection sequence calculation result D5 for a given device U under U (U is an integer) devices. Then, the anomaly factor device calculation unit 350 acquires the anomaly detection sequence assigned to the information about the matched sensor 300 in the anomaly detection sequence calculation result D5 and calculates an anomaly detection sequence total value. Here, the anomaly detection sequence total value is a real number.For example, the anomaly factor device calculation unit 350 sets a representative value as the overall anomaly detection sequence value by using a weighted average of the anomaly detection sequences associated with the information indicating the matched sensor 300. Furthermore, the anomaly factor device calculation unit 350 can set a representative value, such as a minimum or maximum of the anomaly detection sequences associated with the information about the matched sensor 300, as the overall anomaly detection sequence value.

[0341] Subsequently, the Anomaly Factor Device Computing Unit 350 assigns the anomaly detection sequence (hereinafter referred to as the "Device Anomaly Detection Sequence") to Device U in units of the Device U, based on the calculated anomaly detection sequence total value. Here, the Device Anomaly Detection Sequence ouU is a real number. For example, the Anomaly Factor Device Computing Unit 350 assigns the Device Anomaly Detection Sequence to Device U such that the corresponding Device Anomaly Detection Sequence is in ascending order, starting with the lowest Anomaly Detection Sequence total value. Furthermore, if the calculated Anomaly Detection Sequence total values ​​are the same for a majority of devices, the Anomaly Factor Device Computing Unit 350 assigns the same Device Anomaly Detection Sequence to the majority of devices.

[0342] The anomaly factor device calculation unit 350 generates the device anomaly detection sequence calculation result D42, which is information in which the information specifying a device, the anomaly detection sequence total value and the device anomaly detection sequence in units of the device are mapped to each other, and causes the device anomaly detection sequence calculation result D42 to be stored in the data storage unit 20.

[0343] Furthermore, the anomaly factor device calculation unit 350 converts the anomaly propagation sequence calculation result D6 into an anomaly propagation sequence calculation result (hereinafter referred to as a "device anomaly propagation sequence calculation result") D43 in units of the device based on the information D41 from sensors attached to a device and the anomaly propagation sequence calculation result D6.

[0344] In particular, assuming that U devices are represented by devices U (U = 1, 2, ..., U), the anomaly factor device calculation unit 350 compares information indicating sensor 300, which is assigned in the information D41 of sensors attached to a device, with information indicating sensor 300, which is included in the anomaly propagation sequence calculation result D6 for a given device U. Then, the anomaly factor device calculation unit 350 obtains the anomaly propagation sequence associated with the information indicating the matched sensor 300 in the anomaly propagation sequence calculation result D6 and calculates an anomaly propagation sequence total value. Here, the anomaly propagation sequence total value is a real number.For example, the anomaly factor device calculation unit 350 sets a representative value as the overall anomaly propagation sequence value by using a weighted average of the anomaly propagation sequences associated with the information displayed by the matched sensor 300. Furthermore, the anomaly factor device calculation unit 350 can use a representative value, such as a minimum or maximum of the anomaly propagation sequences associated with the information about the matched sensor 300, as the overall anomaly propagation sequence value.

[0345] The Anomaly Factor Device Computing Unit 350 then assigns the anomaly propagation sequence (hereinafter referred to as the "Device Anomaly Propagation Sequence") in device units to Device U based on the calculated anomaly propagation sequence total value. The anomaly detection sequence is a real number. For example, the Anomaly Factor Device Computing Unit 350 assigns the Device Anomaly Propagation Sequence to Device U such that the corresponding anomaly propagation sequence proceeds in ascending order, starting with the lowest calculated anomaly propagation sequence total value. Furthermore, if the calculated anomaly propagation sequence total values ​​are the same in a plurality of devices, the Anomaly Factor Device Computing Unit 350 assigns the same Device Anomaly Propagation Sequence to the plurality of devices.

[0346] The anomaly factor device calculation unit 350 generates the device anomaly propagation sequence calculation result D43, which relates information specifying a device, an anomaly detection device flag, the total anomaly propagation sequence value, and the device anomaly propagation sequence, in units of the device, and causes the device anomaly propagation sequence calculation result D43 to be stored in the data storage unit 20. The anomaly detection device flag indicates whether or not an anomaly detection sensor is present among the sensors 300 that are present in the device in units of the device. The anomaly detection device flag is a Boolean value.

[0347] Furthermore, the device in the first embodiment described above can be equipped with a plurality of sensors 300. For example, if at least one anomaly detection sensor from the plurality of sensors 300 is present in a particular device U, the anomaly factor device calculation unit 350 (true) sets the anomaly detection device flag corresponding to the particular device U in the device anomaly propagation sequence calculation result D43. That is, for example, in a case where a plurality of sensors 300 is provided in a particular device U and there are one or more anomaly detection sensors among the plurality of sensors 300, the anomaly factor device calculation unit 350 (true) sets the anomaly detection device flag according to the particular device U.On the other hand, for example, if a plurality of sensors 300 are provided in a particular device U and there is no anomaly detection sensor among the plurality of sensors 300, the anomaly factor device calculation unit 350 sets (False) for the anomaly detection device flag.

[0348] In addition, the anomaly factor device calculation unit 350 calculates the factor of the anomaly in units of the device based on the generated device anomaly detection sequence calculation result D42 and device anomaly propagation sequence calculation result D43.

[0349] In particular, the anomaly factor device calculation unit 350 calculates the device anomaly factor value for each device from the anomaly detection sequence specified in the device anomaly detection sequence calculation result D42 and the device anomaly propagation sequence specified in the device anomaly propagation sequence calculation result D43. Here, the device anomaly factor value is a real number.

[0350] For example, the Anomaly Factor Device Calculation Unit 350 sets a representative value as the device anomaly factor value by using the weighted average of the device anomaly detection sequence and the device anomaly propagation sequence for each device. Furthermore, the Anomaly Factor Device Calculation Unit 350 can, for example, set a representative value such as a maximum or minimum of the device anomaly detection sequence and the device anomaly propagation sequence as the device anomaly factor value for each device. In addition, if only one of the device anomaly detection sequence and the device anomaly propagation sequence is specified, the Anomaly Factor Device Calculation Unit 350 only needs to set that specified value as the device anomaly factor value.In this case, the anomaly factor device calculation unit 350 can weight the anomaly factor value taking into account the fact that only one order is specified.

[0351] The anomaly factor device calculation unit 350 then assigns an anomaly factor sequence (hereinafter referred to as the "device anomaly factor sequence") to each device based on the anomaly factor value calculated for each device. The device anomaly factor sequence is a real number. For example, the anomaly factor device calculation unit 350 assigns the device anomaly factor sequence to device U such that the corresponding device anomaly factor sequence is in ascending order, starting with the device with the smallest value of the corresponding anomaly factor. Furthermore, if the calculated device anomaly factor values ​​are the same among a plurality of devices U, the anomaly factor device calculation unit 350 assigns the same device anomaly factor sequence to the plurality of devices U.

[0352] The anomaly factor device calculation unit 350 generates the device anomaly factor sequence calculation result D44, which is information that specifies a device, an anomaly detection device flag, a device anomaly factor value, and a device anomaly factor sequence in device units, and causes the device anomaly factor calculation result D44 to be stored in the data storage unit 20. Furthermore, the anomaly factor device calculation unit 350 only needs the value of the anomaly detection device flag for device U, which is set in association with the information that specifies device U in the device anomaly propagation sequence calculation result D43, to determine the device anomaly factor sequence calculation result D44.

[0353] The device unit anomaly factor processing by the anomaly factor device calculation unit 350, as described above, is described with reference to the drawings as a specific example.

[0354] Fig. Figure 24 is a representation describing a concept of an example of a device unit anomaly factor calculation that calculates a factor of an anomaly in units of the device, wherein the process is carried out by the anomaly factor device calculation unit 350 based on the information D41 from sensors attached to a device, the anomaly detection sequence calculation result D5 and the anomaly propagation sequence calculation result D6 in a case where the anomaly factor calculation device 100 according to the first embodiment contains the anomaly factor device calculation unit 350.

[0355] In this example, the number of sensors 300 is set to six, and sensor 300 is represented by sensor Xn (n = 1 to 6). As a further example, the number of devices is three, and the devices are represented by devices U (U = 1, 2, 3). It is also assumed that device 1 is equipped with sensor X1 and sensor X2, device 2 with sensor X3, sensor X4, and sensor X5, and device 3 with sensor X6.

[0356] Based on the anomaly detection sequence (specified by D5C in Fig. 24), which is specified in the anomaly detection sequence calculation result D5 and the information D41 from sensors attached to a device, the anomaly factor device calculation unit 350 calculates an anomaly detection sequence total value suU (specified by D42B in Fig. 24) for each device U, for example by using an average. For example, since in Device 1 the sensors Xn added to Device 1 are sensor X1 and sensor X2, the anomaly factor device calculation unit 350 sets an anomaly detection sequence total value su1 for Device 1 as the mean of the anomaly detection sequence o1 and the anomaly detection sequence o2. Furthermore, for example, in Device 2 the sensors Xn added to Device 2 are sensor X3, sensor X4, and sensor X5. However, of sensors X3, X4, and X5, only sensor X4 is the anomaly detection sensor, i.e., the sensor Xn included in the anomaly detection sequence calculation result D5.Thus, the anomaly factor device calculation unit 350 sets the anomaly detection sequence o4, corresponding to sensor X4, as the anomaly detection sequence total value su2, which corresponds to device 2 as it is. Similarly, the anomaly factor device calculation unit 350 sets the anomaly detection sequence o6, corresponding to sensor X6, as the anomaly detection sequence total value su6, which corresponds to device 3 as it is.

[0357] The anomaly factor device calculation unit 350 then assigns the anomaly detection sequence ouU to each device based on the anomaly detection sequence total value suU calculated for each device. Specifically, the anomaly factor device calculation unit 350 assigns anomaly detection sequences ou1, ou2, and ou3 to devices 1, 2, and 3 respectively, based on the calculated anomaly detection sequence total values ​​su1, su2, and su3.

[0358] Then the anomaly factor device calculation unit 350 generates the device anomaly detection sequence calculation result D42, in which information specifying devices 1, 2 and 3 (specified by D42A in Fig. 24), the anomaly detection sequence totals su1, su2 and su3 (specified by D42B in Fig. 24) and the device anomaly detection sequences ou1, ou2 and ou3 (indicated by D42C in Fig. 24) are assigned to each other, and causes the device anomaly detection sequence calculation result D42 to be stored in the data storage unit 20.

[0359] Furthermore, the anomaly factor device calculation unit 350 calculates an anomaly propagation sequence total value suU (specified by D43C in Fig. 24) for each device, for example by taking an average value based on the anomaly propagation order on (specified by D6C in Fig. 24), which is determined in the anomaly propagation sequence calculation result D6, and which uses information D41 from sensors attached to a device. For example, since in Device 1 the sensors Xn added to Device 1 are sensor X1 and sensor X2, the anomaly factor device calculation unit 350 determines the total anomaly propagation sequence value su1 according to Device 1 as the mean of the anomaly propagation sequence o1 and the anomaly propagation sequence o2. Furthermore, since the sensors Xi added to device 2 are sensor X3, sensor X4 and sensor X5, the anomaly factor device calculation unit 350 in device 2 sets the anomaly propagation sequence total value su2 corresponding to device 2 as the mean of the anomaly propagation sequence o3, the anomaly propagation sequence o4 and the anomaly propagation sequence o5.Similarly, the anomaly factor device calculation unit 350 sets the anomaly propagation sequence o6 corresponding to sensor X6 as the anomaly propagation sequence total value su6 corresponding to device 3 as it is.

[0360] The anomaly factor device calculation unit 350 then assigns the device anomaly propagation sequence ouU based on the total anomaly propagation sequence value suU calculated for each device. Specifically, the anomaly factor device calculation unit 350 assigns the device anomaly propagation sequences ou1, ou2, and ou3 to devices 1, 2, and 3, respectively, based on the calculated total anomaly propagation sequence values ​​su1, su2, and su3.

[0361] Then the anomaly factor device calculation unit 350 generates the device anomaly propagation sequence calculation result D43, in which the information specifying devices 1, 2 and 3 (specified by D43A in Fig. 24), the anomaly detection device flag (specified by D43B in Fig. 24), the anomaly propagation sequence totals su1, su2 and su3 (indicated by D43C in Fig. 24) and the device anomaly propagation sequences ou1, ou2 and ou3 (indicated by D43D in Fig. 24) are assigned to each other, and causes the device anomaly propagation sequence calculation result D43 to be stored in the data storage unit 20.

[0362] Furthermore, the anomaly factor device calculation unit 350 calculates the device anomaly factor value suU using the mean for each device based on the device anomaly detection sequence ouU, which is specified in the device anomaly detection sequence calculation result D42, and the device anomaly propagation sequence ouU, which is specified in the device anomaly propagation sequence calculation result D43. For example, the anomaly factor device calculation unit 350 calculates the device anomaly factor value su1 for device 1 as the mean of the device anomaly detection sequence ou1 and the device anomaly propagation sequence ou1 of device 1.

[0363] The anomaly factor device calculation unit 350 then assigns the device anomaly factor sequence ouU to each device based on the device anomaly factor value suU. For example, the anomaly factor device calculation unit 350 assigns the corresponding device anomaly factor sequences ou1, ou2, and ou3 to devices 1, 2, and 3 respectively, based on the calculated device anomaly factor values ​​su1, su2, and su3.

[0364] The anomaly factor device calculation unit 350 generates the device anomaly factor sequence calculation result D44, which represents information in which the information that devices 1, 2 and 3 specify (specified by D44A in Fig. 24), the anomaly detection device flag (specified by D44B in Fig. 24), the device anomaly factor values ​​su1, su2 and su3 (specified by D44C in Fig. 24) and the device anomaly factor sequences ou1, ou2 and ou3 (specified by D44D in Fig. 24) are assigned to each other, and causes the device anomaly factor sequence calculation result D44 to be stored in the data storage unit 20.

[0365] As described above, when the anomaly factor device calculation unit 350 causes the device anomaly detection sequence calculation result D42, the device anomaly propagation sequence calculation result D43, and the device anomaly factor sequence calculation result D44 to be stored in the data storage unit 20, the anomaly factor calculation result output unit 70 acquires the device anomaly factor detection sequence calculation result D42, the device anomaly propagation sequence calculation result D43, and the device anomaly factor sequence calculation result D44, which are output by the anomaly factor device calculation unit 350 via the data storage unit 20, and provides information regarding the calculation result of the anomaly factor in device units based on the device anomaly detection sequence calculation result D42.of the device anomaly propagation sequence calculation result D43 and the device anomaly factor sequence calculation result D44.

[0366] In particular, the anomaly factor calculation result output unit 70, based on the device anomaly detection sequence calculation result D42, the device anomaly propagation sequence calculation result D43, and the device anomaly factor sequence calculation result D44, outputs information to the display device 400 (hereinafter referred to as "anomaly factor device calculation result display information") to display a screen (hereinafter referred to as "anomaly factor device calculation result screen") that shows the information regarding the calculation result of the anomaly factor for each device by the anomaly factor device calculation unit 350.

[0367] Fig. Figure 25 is a representation showing a screen example of an anomaly factor device calculation result screen displayed on the display device 400 by the anomaly factor calculation result output unit 70 in the first embodiment.

[0368] Fig. Figure 25 shows an example of the anomaly factor device calculation result screen when the number of devices is 3 (devices U. U = 1 to 3).

[0369] In Fig. 25, the anomaly factor device calculation result screen is marked by “D48-1”.

[0370] As in Fig. As shown in 25, the anomaly factor device calculation result screen comprises, for example, ten display frames, namely a display frame D48A, a display frame D48B, a display frame D48C, a display frame D48D, a display frame D48E, a display frame D48F, a display frame D48G, a display frame D48H, a display frame D48I, and a display frame D48J.

[0371] For example, the anomaly factor calculation result output unit 70 causes an anomaly factor device calculation result list to be generated, which lists the information about the calculation result of the anomaly factor in device units for display on the anomaly factor device calculation result screen. For example, the anomaly factor device calculation result list is a list that displays the information specifying the device U, the information specifying the anomaly detection device flag, the device anomaly detection sequence, the device anomaly propagation sequence, the device anomaly factor value, and the device anomaly factor sequence in relation to each other for each device U. In the Fig. The anomaly factor device calculation results screen shown in 25 displays the anomaly factor device calculation results list by “D48-1a”.

[0372] The anomaly factor calculation result output unit 70 outputs anomaly factor device calculation result display information to the display device 400, which causes the information indicating device U of device anomaly factor sequence calculation result D44 to be displayed in display frame D48A, causes the information indicating the anomaly detection device flag of device anomaly propagation sequence calculation result D43 to be displayed in display frame D48B, causes the device anomaly detection sequence of device anomaly detection sequence calculation result D42 to be displayed in display frame D48C, causes the device anomaly propagation sequence of device anomaly propagation sequence calculation result D43 to be displayed in display frame D48D, causesthat the device anomaly factor value of the device anomaly factor sequence calculation result D44 is displayed in display frame D48E, causes the device anomaly factor sequence of the device anomaly factor sequence calculation result D44 to be displayed in display frame D48F, causes sort buttons for reordering, in ascending order, the arrangement order of the anomaly factor device calculation result list based on the device anomaly detection sequence of the device anomaly detection sequence calculation result D42, the device anomaly propagation sequence of the device anomaly propagation sequence calculation result D43 and the device anomaly factor sequence of the device anomaly factor sequence calculation result D44 to be displayed in display frames D48H, D48I and D48J respectively, and causes a checkbox for receiving an instruction,that only one anomaly detection device is displayed in the display frame D48G. As a result, display device 400 shows an anomaly factor device calculation result screen, as shown in , Fig. 25 shown.

[0373] The anomaly factor device calculation results screen, as shown in Fig. 25 is obtained by accessing the anomaly factor calculation results screen, which is already referenced in relation to Fig. 8, is displayed in units of the device, and the functions of the sort button and the checkbox are similar to the functions of the sort button and the checkbox already described with reference to Fig. 8 have been described, so a duplicate description is unnecessary.

[0374] In this case, the anomaly factor device calculation unit 350 continues to operate the anomaly factor calculation device 100, as described in the flowchart of Fig. 13 describes the device unit anomaly factor calculation processing to calculate the factor of the anomaly in device units based on the information D41 from sensors attached to a device, the anomaly detection sequence calculation result D5 and the anomaly propagation sequence calculation result D6 before the process of step ST5 or after the process of step ST5.

[0375] Furthermore, the anomaly factor calculation result output unit 70 can, for example, choose whether to output the information regarding the calculation result of the anomaly factor on a sensor-by-sensor basis or the information regarding the calculation result of the anomaly factor in units of the device, as described in the first embodiment.

[0376] As described above, the anomaly factor calculation device 100 can include the anomaly factor calculation unit 350, which calculates the anomaly factor in device units based on information from sensors attached to a device, the anomaly detection sequence estimated by the anomaly detection sequence calculation unit 40, and the anomaly propagation sequence estimated by the anomaly propagation path tracking unit 50. Thus, the anomaly factor calculation device 100 can enable the operator to efficiently determine the device that caused the anomaly. Furthermore, the anomaly factor calculation device 100 can enable the operator to efficiently record the sequence of inspections to be performed on the devices where the anomaly occurred. <modifikation>

[0377] In the first embodiment described above, the anomaly factor calculation device 100 can include a reference structure diagram output unit 360, which outputs information (hereinafter referred to as "reference structure diagram display information") to the display device 400 for displaying the diagram that relates to the reference structure D2 stored in the data storage unit 20.

[0378] Fig. Figure 26 is a block diagram illustrating a configuration example of the anomaly factor calculation device 100, which contains and is set up to output the reference structure diagram output unit 360 to a display device 400 in the first embodiment.

[0379] Furthermore, the anomaly factor calculation device comprises 100, which are in Fig. Figure 26, which is not shown for the sake of simplicity, includes, in addition to the reference structure diagram output unit 360 and the data storage unit 20, the sensor data acquisition unit 10, the anomaly detection unit 30, the anomaly detection sequence calculation unit 40, the anomaly propagation path tracking unit 50, the anomaly factor calculation unit 60, the anomaly factor calculation result output unit 70, and the control unit. Furthermore, the anomaly factor calculation device 100 does not necessarily have to include the anomaly factor calculation result output unit 70. Moreover, the anomaly factor calculation device 100 is connected to the learning device 200, although this is not shown for the sake of simplicity. Fig. 26 is not shown.

[0380] Based on the reference structure D2, the anomaly detection sensor information D3, and the anomaly factor calculation result D7, which are stored in the data storage unit 20, the reference structure diagram output unit 360 outputs the reference structure diagram display information to the display device 400 to display a diagram that relates to the reference structure D2. The diagram or graph that relates to the reference structure D2 is, for example, a diagram in which the reference structure D2, the anomaly detection sensor, and the calculation result of the anomaly factor are mapped to each other.

[0381] The display device 400 displays a screen (hereinafter referred to as the "diagram screen") on which a diagram related to the reference structure D2 is displayed based on the reference structure diagram display information output by the reference structure diagram output unit 360.

[0382] Furthermore, the reference structure diagram output unit 360 can be provided in a device that is connected to the anomaly factor calculation device 100 via a wired or wireless signal line outside the anomaly factor calculation device 100, such as the display device 400.

[0383] An example of a diagram screen displayed on the display device 400 by the reference structure diagram output unit 360, which outputs the reference structure diagram display information in a case where the anomaly factor calculation device 100 contains the reference structure diagram output unit 360, is described with reference to the drawings.

[0384] Fig. Figure 27 is a representation to describe an example of a diagram screen displayed on the display device 400 by the reference structure diagram output unit 360, which outputs the reference structure diagram display information in a case where the anomaly factor calculation device 100 includes the reference structure diagram output unit 360 in the first embodiment.

[0385] In this example, the number of sensors 300 is set to six, and sensor 300 is represented by a sensor Xn (n = 1 to 6).

[0386] Furthermore, an example of the in Fig. The diagram screen shown in section 27 is an example of a diagram screen displayed based on the reference structure diagram display information output by the reference structure diagram output unit 360 in a case where the content of the reference structure D2 stored in the data storage unit 20 is the same as the one shown in the diagram screen. Fig. The content shown in section 28 is the content of the anomaly detection sensor information D3 of the in Fig. The content shown in section 29 is the anomaly factor calculation result D7, which is in Fig. The content shown is 30.

[0387] Furthermore, when outputting the reference structure diagram display information, the reference structure diagram output unit 360 determines whether each element of the reference structure D2 is an element with a dependency relationship or an element without a dependency relationship for each statistical index, and transforms the reference structure D2 into a reference structure (hereinafter referred to as "reference structure after determination of dependency relationship") by using information that indicates the presence or absence of a dependency relationship as an element.

[0388] For example, the Reference Structure Diagram output unit 360 determines, for each statistical index, that each element of the reference structure D2 is an element with a dependency relationship if each element is equal to or greater than a preset dependency selection threshold, and determines that each element is an element without a dependency relationship if each element is less than the dependency selection threshold. Then, for example, the Reference Structure Diagram output unit 360 generates the reference structure after determining the dependency relationship, which is represented by a matrix where an element with a dependency relationship is "1" and an element without a dependency relationship is "0".

[0389] In Fig. 28 is a dependency-degree-of-determination reference structure for each statistical index (in Fig. 28 (labeled by D2R) together with the reference structure D2.

[0390] In Fig. 27. The chart screen is labeled D45. For example, a reference structure diagram (in Fig. 27 (labeled D45-1) is shown, which is a directed diagram in which the reference structure D2 is represented by the sensor Xn as a node and the dependency relationship between the sensors Xn is represented by an edge.

[0391] In addition to the reference structure diagram, a screen (hereinafter referred to as the "index label screen") is displayed on the diagram screen (in Fig. 27 (marked by D45I) displays a checkbox (hereinafter referred to as the "index determination checkbox") to specify a type of statistical index as the target for displaying a corresponding element in the reference structure diagram. Since there are three types of statistical indices, the following is displayed on the Fig. The index label screen shown in image 27 displays an index label checkbox labeled "Statistical Index 1" for entering the name of a first type of statistical index, an index designation checkbox labeled "Statistical Index 2" for entering the name of a second type of statistical index, and an index designation checkbox labeled "Statistical Index 3" for entering the name of a third type of statistical index. For example, the operator specifies a statistical index for which the correlation (edge) between the corresponding sensors Xn should be displayed by activating the index designation checkbox. Fig. Figure 27 shows a state in which the index determination checkbox corresponding to "statistical index 1" and the index determination checkbox corresponding to "statistical index 2" are selected on the index determination screen; in other words, a state in which the first and second statistical indices are designated. Therefore, only the edges corresponding to the first and second types of statistical indices, which are the statistical indices for which the index determination checkbox is selected, are displayed in the reference structure diagram.

[0392] For example, as in Fig. Figure 27 shows the edges corresponding to the respective statistical indices, represented with different line types, so that it is clear which statistical index the edge corresponds to. In the Fig. In the reference structure diagram shown in Figure 27, the edge corresponding to the first type of statistical index is represented by a solid arrow (see, for example, D45G), and the edge corresponding to the second type of statistical index is represented by a dotted arrow (see, for example, D45H). Furthermore, this is only an example, and the edge corresponding to each statistical index could, for instance, be represented by a different arrow color.

[0393] Furthermore, a screen (hereinafter referred to as the "node condition label screen") is displayed on the diagram screen (in Fig. 27 (marked by D45J) to display a checkbox (hereinafter referred to as the "node condition label checkbox") for labeling a display condition (hereinafter referred to as the "node display condition") with respect to a node. The node display condition is preset. In the Fig. On the node condition label screen shown in Figure 27, three conditions are defined as node display conditions: "Show anomaly detection sensor only," "Highlight anomaly detection sensor," and "Show anomaly factor order." The operator designates a node display condition, for example, by selecting a checkbox for the node condition label checkbox. Fig. Figure 27 illustrates a state in which the checkbox corresponding to "Highlight anomaly detection sensor" and the checkbox corresponding to "Show anomaly detection sequence" are enabled on the node condition label screen.

[0394] In the reference structure diagram, the nodes (in Fig. 27 (represented as D45A, D45D, D45E, and D45F), which correspond to sensors X1, X4, X5, and X6, which are anomaly detection sensors, are filled in and displayed. Furthermore, it is assumed here that the anomaly detection sensor is highlighted by filling in and displaying the node corresponding to the anomaly detection sensor, but the method for highlighting the anomaly detection sensor is not limited to this, and the node corresponding to the anomaly detection sensor can be highlighted by another method.

[0395] Furthermore, the reference structure diagram displays the anomaly factor sequence at the node corresponding to sensor Xn. In the Fig. In the reference structure diagram shown in Figure 27, the anomaly factor order is displayed as "Rank 1", "Rank 2", "Rank 3", "Rank 4", or "Rank 5". The anomaly factor order is also displayed as "Rank Δ", but the display method for the anomaly factor order is not necessarily limited to this; it simply needs to be displayed in a way that makes the anomaly factor order understandable.

[0396] Furthermore, the three conditions “show only anomaly detection sensor”, “highlight anomaly detection sensor” and “show anomaly factor order” are set here as node display conditions, but this is just an example, and other conditions can be set as node display conditions.

[0397] In the reference structure diagram, for example, a sensor name is displayed at each node to identify sensor Xn for the operator who has checked the diagram screen. Fig. The sensor names “X1”, “X2”, “X3”, “X4”, “X5” and “X6” are displayed.

[0398] The edge is displayed, for example, when the statistic, which is an element of the reference structure D2, is greater than the threshold for selecting dependencies, which is specified for each statistical index. If the statistic is greater than the threshold, this means that a dependency relationship exists between the sensors Xn. Furthermore, after determining the dependency relationship, the reference structure diagram output unit 360 can determine the dependency relationship between the sensors Xn from the reference structure.

[0399] If the dependency relationship between sensors Xn is unidirectional, the directed diagram displays an edge of a one-sided arrow, and if the dependency relationship between sensors Xn is bidirectional, the directed diagram displays an edge with a two-sided arrow. For example, since there is a one-directional dependency relationship from sensor X2 to sensor X1, as in Fig. As shown in 27, an edge (in the directed diagram) is formed. Fig. 27 marked by D45G) of a one-sided arrow from the node that connects sensor X2 (in Fig. 27 marked by D45B) indicates, to the node that displays sensor X1 (in Fig. 27 (labeled by D45A) is shown. Since sensor X2 and sensor X6 have a bidirectional dependency relationship, as shown in Fig. 27 shows an edge (in Fig. 27 marked by D45H) of a double-sided arrow between the node indicating sensor X2 and the node indicating sensor X6 (in Fig. 27 (labeled D45F) is shown in the directed diagram.

[0400] Fig. Figure 31 is a flowchart to describe an example of the operation of the anomaly factor calculation device 100 in a case in which the anomaly factor calculation device 100 comprises the reference structure diagram output unit 360 in the first embodiment.

[0401] The anomaly factor calculation device 100 performs the calculation shown in the flowchart of Fig. The process shown in section 31, in addition to the one described in the flowchart of Fig. The process described in section 13 is further assumed to be the one described in the flowchart of Fig. 31. The process shown is performed at least once after processing steps ST1 to ST5. Fig. 13 is carried out. The one in the flowchart of Fig. The process shown in step 31 can, for example, be carried out after the processing of step ST5 of Fig. 13, before processing step ST6, after processing step ST6 or in parallel with processing step ST6.

[0402] The reference structure diagram output unit 360 receives a display instruction for the reference structure diagram (step ST31).

[0403] The operator uses an input device, such as a mouse or keyboard, to call up an input screen for a display instruction for the reference structure diagram on the display device 400. The operator enters a display instruction for the reference structure diagram via the input screen. The reference structure diagram output unit 360 receives a display instruction for the reference structure diagram entered by the operator.

[0404] Based on the reference structure D2 stored in the data storage unit 20, the anomaly detection sensor information D3, and the anomaly factor sequence calculation result D7, the reference structure diagram output unit 360 outputs the reference structure diagram display information to the display device 400 to display a diagram that relates to the reference structure D2 (step ST32). For example, a diagram screen is displayed on the display device 400, as shown in Fig. 27 shown, displayed.

[0405] As described above, the anomaly factor calculation device 100 includes the reference structure diagram output unit 360, which allows the anomaly factor calculation device 100 to improve the explainability of information about the anomaly factor calculation result. <modifikation>

[0406] In the first embodiment described above, the learning device 200 can include a learning sensor pair generation unit 370, which generates a pair of sensors 300 from the plurality of sensors 300 based on the connection relationship between the plurality of devices forming the target plant and information (hereinafter referred to as "plant design information") that defines the plurality of sensors 300 in the plurality of devices. In this case, the reference structure learning unit 240 acquires the learning sensor data generated by the learning sensor pair generation unit 370 for each sensor pair 300 and learns the reference structure D2.Furthermore, the operator or the like generates the plant design information on the basis of a design drawing and enters the plant design information, which is generated by operating the input device, such as a mouse or a keyboard, and the learning device 200 acquires the plant design information by receiving the entered plant design information.

[0407] Fig. Figure 32 is a block diagram showing a configuration example of the learning device 200 including the learning sensor pair generation unit 370 in the first embodiment.

[0408] Furthermore, the learning device includes 200 even if these are in Fig. Figure 32, which is not shown for the sake of simplicity, includes the learning sensor data acquisition unit 210 and the learning preprocessing unit 230, in addition to the learning sensor pair generation unit 370, the reference structure learning unit 240, and the learning data storage unit 220. Furthermore, the learning device 200 is connected to the anomaly factor calculation device 100, although for the sake of simplicity it is shown in Figure 32. Fig. 32 is not shown.

[0409] The learning sensor pair generation unit 370 acquires the plant design information and generates a pair of sensors 300, which are used when the reference structure learning unit 240 learns the reference structure D2 from the plurality of sensors 300 based on the acquired plant design information.

[0410] Specifically, the learning sensor pair generation unit 370 determines a combination of the two sensors 300 based on the plant design information and generates information (hereinafter referred to as "sensor pair information") D47 listing the combinations.

[0411] The learning sensor pair generation unit 370 outputs the generated sensor pair information D47 to the reference structure learning unit 240.

[0412] When the sensor pair information D47 is output by the learning sensor pair generation unit 370, the reference structure learning unit 240 calculates a statistic that shows a relationship between two different parts of sensor data for a variety of parts of sensor data contained in the learning data D23, based on the pair of sensors 300 set in the sensor pair information D47, and learns the reference structure D2 based on the calculated statistic.

[0413] The processing, in which the learning sensor pair generation unit 370 generates the sensor pair information D47, is described with reference to the drawings using a specific example.

[0414] Fig. Figure 33 is a diagram to describe a concept of an example of a method for generating the sensor pair information D47 based on the plant design information D46 by the learning sensor pair generation unit 370 in a case in which the learning device 200 includes the learning sensor pair generation unit 370 in the first embodiment.

[0415] In this example, the number of sensors 300 is set to eight, and sensor 300 is represented by a sensor Xn (n = 1 to 8).

[0416] Furthermore, it is assumed here, by way of example, that the target system comprises five devices (a device D46A, a device D46B, a device D46C, a device D46D and a device D46E).

[0417] Furthermore, it is assumed that the device D46A is equipped with sensor X1 and sensor X2, the device D46B with sensor X3, the device D46C with sensor X4 and sensor X5, the device D46D with sensor X6 and sensor X7, and the device D46E with sensor X8.

[0418] Furthermore, it is assumed that device D46A is related to device D46B, device D46B to devices D46A, D46C and D46D, device D46C to devices D46B and D46E, device D46D to device D46B and device D46E to device D46C. Additionally, devices that are related in the design are connected by a non-directional line. In this case, the plant design information D46 has the one in Fig. The content shown is in section 33. It should be noted that in Fig. Figure 33 shows, as an example, the plant design information D46 in a block diagram, but this is only an example. The plant design information D46 can be any information as long as a connection relationship is known between a plurality of devices that constitute the target plant and a plurality of sensors Xn that are present in the plurality of devices.

[0419] The learning sensor pair generation unit 370 generates a pair with two different sensors Xn based on the plant design information D46. In particular, the learning sensor pair generation unit 370 generates a pair comprising a sensor Xn that has been added to a particular device and a sensor Xn that has been added to a device and has a connection relationship with the device.

[0420] In the Fig. In the example shown in 33, the device D46A and the device D46B are connected according to the plant design information D46. In this case, the sensor X1 and the sensor X2 provided in the device D46A, as well as the sensor X3 provided in the device D46B, are also connected. Therefore, the learning sensor pair generation unit 370 generates a pair of sensor X1 and sensor X3 and a pair of sensor X2 and sensor X3.

[0421] Furthermore, in a case where two or more sensors Xn are present in a device, the learning sensor pair generation unit 370 generates a pair of different sensors Xn among the two or more sensors Xn that are present in the same device.

[0422] In the Fig. In the example shown in Figure 33, the device D46A is equipped with sensor X1 and sensor X2 according to the plant design information D46. Therefore, the learning sensor pair generation unit 370 generates a pair from sensor X1 and sensor X2.

[0423] As described above, the learning sensor pair generating unit 370, for example, generates a pair with two sensors present in different interconnected devices, and a pair with two different sensors present in one device, as a pair of sensors.

[0424] Furthermore, the sensor pair generated by the learning sensor pair generation unit 370, as described above, is merely an example, and the learning sensor pair generation unit 370 can, for example, only generate a pair comprising two sensors provided in different devices that are in a connection relationship to each other, or can only generate a pair comprising two different sensors provided in one device.

[0425] As described above, in a case where the learning device 200 contains the learning sensor pair generation unit 370, the learning device 200 operates according to the flowchart of Fig. As described in section 14, the sensor pair information D47 is retrieved and output to the reference structure learning unit 240 before the processing of step ST231 is performed. In step ST231, the reference structure learning unit 240 obtains two different pairs of sensor data based on the training data D23 and the sensor pair information D47. It should be noted that the reference structure learning unit 240 defines all combinations of the parts of sensor data contained in the training data D23 as sensor data pairs based on the sensor pair information D47.

[0426] As described above, the learning device 200 contains the learning sensor pair generation unit 370, which generates a pair of sensors 300 from the plurality of sensors 300 based on the plant design information. The reference structure learning unit 240 is configured to obtain the learning sensor data for each pair of sensors 300 generated by the learning sensor pair generation unit 370 and to learn the reference structure D2. This allows the learning device 200 to suppress the possibility of detecting the dependency relationship between sensors 300 that are of minor relevance to the design and enables the reference structure D2 to learn with improved reliability. Consequently, the learning device 200 can provide the reference structure D2 to the anomaly factor calculation device 100, which is capable of accurately calculating the anomaly factor.

[0427] As described above, the anomaly factor calculation device 100 according to the first embodiment includes the sensor data acquisition unit 10 for acquiring a plurality of parts of time-series sensor data collected from a plurality of sensors 300 provided in a plurality of plant components forming a target plant, the anomaly detection unit 30 for detecting a plurality of anomaly detection sensors in which an anomaly has occurred among a plurality of the sensors 300 based on a plurality of parts of the sensor data acquired by the sensor data acquisition unit 10, and the anomaly detection sequence calculation unit 40 for calculating an anomaly detection sequence in which the occurrence of the anomaly for a plurality of the anomaly detection sensors is determined based on a detection time.to which the anomaly detection unit 30 has detected a plurality of the anomaly detection sensors, wherein the anomaly propagation path tracking unit 50 calculates an anomaly propagation sequence in which the anomaly has propagated based on anomaly detection sensor information D3 with respect to a plurality of the anomaly detection sensors detected by the anomaly detection unit 30, and a calculated structure (reference structure D2) that specifies a dependency relationship between the plant components, and the anomaly factor calculation unit 60 to determine a factor of the anomaly based on the anomaly detection sequence calculated by the anomaly detection sequence calculation unit 40 and the anomaly propagation sequence calculated by the anomaly propagation path tracking unit 50.

[0428] Therefore, the anomaly factor calculation device 100 can estimate the factor of anomalies that have occurred in the facility or plant, regardless of the complexity or size of the facility.

[0429] Furthermore, the anomaly factor calculation device 100 includes the anomaly factor calculation result output unit 70 for outputting information about an anomaly factor calculation result of the factor of the anomaly by the anomaly factor calculation unit 60.

[0430] Therefore, the Anomaly Factor Calculation Device 100 improves the interpretability and explainability of the anomaly factor calculation result for the operator. The Anomaly Factor Calculation Device 100 can reduce unnecessary inspection work by the operator and decrease their workload. Furthermore, the Anomaly Factor Calculation Device 100 can estimate the anomaly factor using a quantitative index that is independent of human subjectivity and provide a rationale for the calculation. The operator can thus establish a test sequence for the system with less effort.

[0431] Furthermore, the anomaly factor calculation device 100 can be set up to detect the anomaly detection sensor using the univariate type anomaly detection method.

[0432] Therefore, the anomaly factor calculation device 100 can better detect an anomaly where only part of sensor data D1 changes.

[0433] Furthermore, the anomaly factor calculation device 100 can be configured to detect the anomaly detection sensor using a multivariate anomaly detection method.

[0434] Therefore, the anomaly factor calculation device 100 can better detect an anomaly in which the relationship between the multitude of parts of sensor data D1 changes.

[0435] Furthermore, the anomaly factor calculation device 100 can be set up to detect the anomaly detection sensor using the univariate type anomaly detection method and the multivariate type anomaly detection method.

[0436] Therefore, the anomaly factor calculation device 100 can better detect an anomaly where only a part of the sensor data D1 changes, or an anomaly where the relationship between the multiple parts of the sensor data D1 changes.

[0437] Furthermore, the anomaly propagation path tracking unit 50 can be set up in the anomaly factor calculation device 100, which calculates the anomaly propagation sequence on the basis of the anomaly detection sensor information D3, the plant operating state information D31, which specifies the operating state of the target plant, and the calculated structure (reference structure D32), which specifies the dependency relationship between the plant components depending on the operating state of the target plant.

[0438] Therefore, the anomaly factor calculation device 100 can cope with a change in the dependency relationship between the sensors 300 due to the change in the operating state in the target plant and accurately calculate the anomaly factor based on the reference structure D32 with improved reliability.

[0439] Furthermore, the anomaly factor calculation device 100 can include the reference structure correction unit 330, which corrects the dependency relationship between the parts of sensor data for the calculated structure (reference structure D2) on the basis of the information about dependent pairs D33, which refer to the pair of sensors that have a dependency relationship between the majority of sensors 300, and the information about non-dependent pairs D34, which refer to the pair of sensors 300 that do not have a dependency relationship.

[0440] Therefore, the anomaly factor calculation device 100 can improve the reliability of the calculated structure and accurately calculate the anomaly factor.

[0441] Furthermore, the anomaly factor calculation device 100 can include the relationship change calculation unit 340 to compare the calculated structure with the calculated structure at the time of the anomaly's occurrence based on the calculated structure (reference structure D2), the calculated structure at the time of the anomaly's occurrence (reference structure D36), and the anomaly detection sensor information D3, and to calculate a change in the relationship among the parts of sensor data. The anomaly factor calculation unit 60 can be configured to calculate an anomaly factor taking into account the change in the relationship between the parts of sensor data, calculated by the relationship change calculation unit 340 based on the anomaly detection sequence calculated by the anomaly detection sequence calculation unit 40 and the anomaly propagation sequence calculated by the anomaly propagation path tracking unit 50.to determine.

[0442] Therefore, the anomaly factor calculation device 100 improves the reliability of the anomaly factor sequence calculation result D7 and can accurately calculate the anomaly factor.

[0443] Furthermore, the anomaly factor calculation device 100 can include an anomaly factor device calculation unit 350 to calculate an anomaly factor in device units based on the information D41 from sensors attached to a device in which a device provided in the target plant and the sensor 300 provided in the device are associated with each other, the anomaly detection sequence calculated by the anomaly detection sequence calculation unit 40 and the anomaly propagation sequence calculated by the anomaly propagation path tracking unit 50.

[0444] Therefore, the anomaly factor calculation device 100 can enable the operator to efficiently determine the device that caused the anomaly. Furthermore, the anomaly factor calculation device 100 can allow the operator to efficiently record the sequence of inspections to be performed on the devices where the anomaly occurred.

[0445] Furthermore, the anomaly factor calculation device 100 can include the reference structure diagram output unit 360 to output the reference structure diagram display information to display a diagram in which the calculated structure, the anomaly detection sensor and a calculation result of the anomaly factor based on the calculated structure (reference structure D2), the anomaly detection sensor information D3 and the information regarding the calculation result of the anomaly factor calculated by the anomaly factor calculation unit 60 are related to each other.

[0446] Therefore, the anomaly factor calculation device 100 can improve the explainability of the information about the calculation result of the anomaly factor.

[0447] As described above, the learning device 200 according to the first embodiment comprises the learning sensor data acquisition unit 210, which acquires as learning data candidates a plurality of parts of sensor data in time series form, collected from the plurality of sensors 300 supplied to the target plant during the time of normal operation of the target plant, and the reference structure learning unit 240, which, using the plurality of learning data candidates acquired by the learning sensor data acquisition unit 210 as a plurality of parts of learning data, calculates at least one statistic between the plurality of parts of learning data on the basis of the learning data and, on the basis of the calculated statistic, learns a calculated structure (reference structure D2) that indicates a dependency relationship between plant components.

[0448] Therefore, the learning device 200 can understandably extract the relevance between the parts of sensor data D1, and consequently, it is possible to provide the reference structure D2, in which the overlooking of the connection relationship of sensor 300 is suppressed. The learning device 200 can cause the anomaly factor calculation device 100 to better track sensor 300, which is the source of the anomaly, and improve the computational accuracy of the anomaly factor by providing the anomaly factor calculation device 100 with the calculated structure (reference structure D2) while tracking sensor 300, which is the source of the anomaly.

[0449] Furthermore, the learning device 200 includes the learning preprocessing unit 230 to acquire a variety of parts of learning data to be used for learning based on a variety of learning data candidates acquired by the learning sensor data acquisition unit 210, and the reference structure learning unit 240 can be configured to compute at least one of the statistics among a variety of parts of learning data based on the learning data acquired by the learning preprocessing unit 230 and to learn the computed structure (reference structure D2) based on the computed statistics.

[0450] Therefore, the learning device 200 can understandably extract the relevance between the parts of sensor data D1, and consequently, it is possible to provide the reference structure D2, in which the overlooking of the connection relationship of sensor 300 is suppressed. The learning device 200 can cause the anomaly factor calculation device 100 to better track sensor 300, which is the source of the anomaly, and improve the computational accuracy of the anomaly factor by providing the anomaly factor calculation device 100 with the calculated structure (reference structure D2) while tracking sensor 300, which is the source of the anomaly.

[0451] Furthermore, the learning device 200 can be equipped with the learning preprocessing unit 230 to select from the multitude of learning data candidates acquired by the learning sensor data acquisition unit 210 a multitude of learning data candidates whose variance is less than the selection threshold, and to acquire the multitude of selected learning data candidates as a multitude of parts of learning data.

[0452] Therefore, the learning device 200 can understandably extract the relevance between the parts of sensor data D1, and consequently, it is possible to provide the reference structure D2 in which the overlooking of the connection relationship of sensor 300 is suppressed. The learning device 200 can cause the anomaly factor calculation device 100 to track sensor 300, which is the source of the anomaly, more appropriately and improve the estimation or calculation accuracy of the anomaly factor by providing the anomaly factor calculation device 100 with the calculated structure (reference structure) while tracking sensor 300, which is the source of the anomaly.

[0453] Furthermore, the reference structure learning unit 240 can be set up in the learning device 200 to calculate the statistics using the waveform-based statistical index.

[0454] Therefore, the learning device 200 can track the anomaly propagation based on the dependency relationship of a similarity in the waveform and provide the calculated structure (reference structure D2) that can more appropriately calculate the factor of the anomalies.

[0455] Furthermore, the learning device 200 can be set up to calculate the statistics using the distribution-based statistical index.

[0456] Therefore, the learning device 200 can track the anomaly propagation based on the dependency relationship of a similarity in the distribution and provide the calculated structure (reference structure D2) with which the anomaly factor can be calculated more appropriately.

[0457] Furthermore, the learning device 200 can be set up to calculate the statistics using the waveform-based statistical index and the distribution-based statistical index.

[0458] Therefore, the learning device 200 can track the anomaly propagation based on the dependency relationship of the similarity of the waveform or distribution and provide the calculated structure (reference structure D2) with which the anomaly factor can be calculated more appropriately.

[0459] Furthermore, the learning device 200 includes the learning sensor pair generation unit 370 to generate a pair of sensors 300 from a plurality of sensors 300 based on a connection relationship between a plurality of devices forming the target plant and the plant design information D46, in which a plurality of sensors 300 provided in a plurality of devices are defined, and the reference structure learning unit 240 can be configured to obtain the learning data based on the pair of sensors 300 generated by the learning sensor pair generation unit 370 and to learn the calculated structure (reference structure D2).

[0460] Therefore, the learning device 200 can suppress the possibility of detecting the dependency relationship between the sensors 300, which has little relevance to the design, and learn the reference structure D2 with improved reliability. As a result, the learning device 200 can provide the reference structure D2 to the anomaly factor calculation device 100, which is then able to accurately calculate the anomaly factor.

[0461] It should be noted that in the present disclosure any component of the embodiment can be modified or any component of the embodiment can be omitted. INDUSTRIAL APPLICABILITY

[0462] In an anomaly factor calculation device according to the present disclosure, an anomaly factor calculation device can calculate a factor of an anomaly that has occurred in a facility or plant, regardless of the complexity or size of the facility or plant. REFERENCE MARK LIST

[0463] 1000: Precise diagnostic system, 100: Anomaly factor calculation device, 10, 310: Sensor data acquisition unit, 20, 320: Data storage unit, 30: Anomaly detection unit, 40: Anomaly detection sequence calculation unit, 50: Anomaly propagation path tracking unit, 60: Anomaly factor calculation unit, 70: Anomaly factor calculation result output unit, 330: Reference structure correction unit, 340: Relationship change calculation unit, 350: Anomaly factor device calculation unit, 360: Reference structure diagram output unit, 200: Learning device, 210: Learning sensor data acquisition unit, 220: Learning data storage unit, 230: Learning preprocessing unit, 240: Reference structure learning unit, 370: Learning sensor pair generation unit, 300: Sensor, 400: Display device, 1601: Processing circuit, 1602: Input interface device, 1603: Output interface device, 1604: Processor, 1605: Memory QUOTES CONTAINED IN THE DESCRIPTION

[0000] This list of documents submitted by the applicant was generated automatically and is included solely for the convenience of the reader. This list is not part of the German patent or utility model application. The DPMA assumes no liability for any errors or omissions. Cited patent literature

[0000] WO 2017 / 159016 A1

[0005] Cited non-patent literature

[0000] Non-Patent Literature: KEOGH, Eamonn; LIN, Jessica; FU, Ada. Hot Wire: Efficiently finding the most unusual subsequence of a time series. In: Data Mining, Fifth International IEEE Conference on Data Mining. IEEE, 2005

[0042] < / modifikation> < / modifikation> < / modifikation> < / modifikation> < / modifikation> < / modifikation> < / modifikation> < / modifikation> < / modifikation> < / bezugsstrukturlernen> < / lerndatenauswahl> < / anomalieausbreitungsreihenfolgeberechnung>

Claims

[1] Anomaly factor calculation device comprising: a sensor data acquisition unit for acquiring a plurality of time-series sensor data acquired by a plurality of sensors provided in a plurality of plant components constituting a target plant; an anomaly detection unit for detecting a plurality of anomaly detection sensors in which an anomaly has occurred among the plurality of sensors, based on a plurality of sensor data acquired by the sensor data acquisition unit; an anomaly detection order calculation unit for calculating an order of anomaly detection in which the occurrence of the anomaly is detected for the plurality of anomaly detection sensors based on a detection time at which the anomaly detection unit detected the plurality of anomaly detection sensors; an anomaly propagation path tracking unit for calculating an anomaly propagation order in which the anomaly has propagated, based on anomaly detection sensor information regarding the plurality of anomaly detection sensors detected by the anomaly detection unit and a calculated structure indicating a dependency relationship between the plant components; and an anomaly factor calculation unit for calculating a factor of the anomaly based on the anomaly detection order calculated by the anomaly detection order calculation unit and the anomaly propagation order calculated by the anomaly propagation path tracking unit. [2] An anomaly factor calculation device according to claim 1, wherein the calculated structure is represented by a matrix. [3] Anomaly factor calculation device according to claim 1 or 2, comprising: an anomaly factor calculation result output unit for outputting information regarding a calculation result of the anomaly factor by the anomaly factor calculation unit. [4] The anomaly factor calculating device according to any one of claims 1 to 3, wherein the anomaly detecting unit detects the anomaly detecting sensors with a univariate anomaly detecting method. [5] The anomaly factor calculating device according to any one of claims 1 to 3, wherein the anomaly detecting unit detects the anomaly detecting sensor with a multivariate anomaly detecting method. [6] The anomaly factor calculating device according to any one of claims 1 to 3, wherein the anomaly detecting unit detects the anomaly detecting sensor using a univariate anomaly detecting method and a multivariate anomaly detecting method. [7] The anomaly factor calculation device according to any one of claims 1 to 6, wherein the anomaly propagation path tracking unit calculates the anomaly propagation order based on the anomaly detection sensor information, the plant operating state information indicating an operating state of the target plant, and the calculated structure indicating a dependency relationship between the plant components depending on the operating state of the target plant. [8] Anomaly factor calculation device according to any one of claims 1 to 7, comprising: a reference structure correction unit for correcting a dependency relationship between the pieces of sensor data for the calculated structure based on dependent pair information relating to a pair of the sensors having a dependency relationship among the plurality of sensors and non-dependent pair information relating to a pair of the sensors having no dependency relationship. [9] Anomaly factor calculation device according to any one of claims 1 to 8, comprising: a relationship change calculation unit for comparing the calculated structure with the calculated structure at a time of occurrence of the anomaly based on the calculated structure, the calculated structure at a time of occurrence of the anomaly and the anomaly detection sensor information, and calculating a change in a relationship between the sensor data, wherein the anomaly factor calculation unit calculates a factor of the anomaly taking into account a change in a relationship between the pieces of sensor data calculated by the relationship change calculation unit based on the anomaly detection order calculated by the anomaly detection order calculation unit and the anomaly propagation order calculated by the anomaly propagation path tracking unit. [10] Anomaly factor calculation device according to any one of claims 1 to 9, comprising: an anomaly factor device calculation unit for calculating a factor of the anomaly in units of the device based on information from sensors attached to the device in which a device provided in the target facility and the sensor provided in the device are associated with each other, the anomaly detection order calculated by the anomaly detection order calculation unit, and the anomaly propagation order calculated by the anomaly propagation path tracking unit. [11] Anomaly factor calculation device according to any one of claims 1 to 10, comprising: a reference structure diagram output unit for outputting reference structure diagram display information for displaying a diagram in which the calculated structure, the abnormality detection sensor, and a calculation result of an abnormality factor are associated with each other based on the calculated structure, the abnormality detection sensor information, and the information regarding the calculation result of the abnormality factor calculated by the abnormality factor calculation unit. [12] Learning device comprising: a learning sensor data acquisition unit for acquiring, as learning data candidates, a plurality of pieces of time-series sensor data acquired by a plurality of sensors provided in a target facility during a period of normal operation of the target facility; and a reference structure learning unit for calculating, using a plurality of pieces of the candidate learning data acquired by the learning sensor data acquiring unit as a plurality of pieces of learning data, at least one of the statistics of the plurality of pieces of learning data based on the plurality of pieces of learning data, and learning a calculated structure indicating a dependency relationship between the plant components based on the calculated statistics. [13] Learning device according to claim 12, comprising: a learning preprocessing unit for acquiring the plurality of pieces of learning data to be used for learning based on the plurality of learning data candidates acquired by the learning sensor data acquiring unit, wherein the reference structure learning unit calculates at least one of the statistics among the plurality of pieces of learning data based on the learning data acquired by the learning preprocessing unit, and learns the calculated structure based on the calculated statistics. [14] The learning apparatus according to claim 13, wherein the learning preprocessing unit selects the plurality of learning data candidates whose variance is smaller than a selection threshold among the plurality of learning data candidates acquired by the learning sensor data acquiring unit, and acquires a plurality of the selected learning data candidates as the plurality of pieces of learning data. [15] The learning apparatus according to any one of claims 12 to 14, wherein the reference structure learning unit calculates the statistics using a waveform-based statistical index. [16] The learning apparatus according to any one of claims 12 to 14, wherein the reference structure learning unit calculates the statistics using a distribution-based statistical index. [17] The learning apparatus according to any one of claims 12 to 14, wherein the reference structure learning unit calculates the statistics using a waveform-based statistical index and a distribution-based statistical index. [18] Learning device according to one of claims 12 to 17, comprising: a learning sensor pair generation unit for generating a pair of sensors from the plurality of sensors based on a connection relationship between a plurality of devices constituting the target plant and plant design information in which the plurality of sensors provided in a plurality of the devices are defined, wherein the reference structure learning unit acquires the learning data based on the sensor pair generated by the learning sensor pair generation unit and learns the calculated structure. [19] Precise diagnostic system that includes: the anomaly factor calculation device according to any one of claims 1 to 11; and the learning device according to one of claims 12 to 18. [20] Anomaly factor calculation method, which includes: causing a sensor data acquisition unit to acquire a plurality of time-series sensor data acquired by a plurality of sensors provided in a plurality of plant components constituting a target plant; causing an anomaly detection unit to detect a plurality of anomaly detection sensors in which an anomaly has occurred among the plurality of sensors based on the plurality of sensor data acquired by the sensor data acquisition unit; causing an anomaly detection order calculation unit to calculate an anomaly detection order in which the occurrence of the anomaly is detected for the plurality of anomaly detection sensors based on a detection time at which the anomaly detection unit detected the plurality of anomaly detection sensors; Causing an anomaly propagation path tracking unit to calculate an anomaly propagation order in which the anomaly has propagated based on anomaly detection sensor information regarding the plurality of anomaly detection sensors detected by the anomaly detection unit and a calculated structure indicating a dependency relationship between the plant components; and Causing an anomaly factor calculation unit to calculate a factor of the anomaly based on the anomaly detection order calculated by the anomaly detection order calculation unit and the anomaly propagation order calculated by the anomaly propagation path tracking unit.

Images