Abnormality diagnosis device, abnormality diagnosis method, and program

Abstract

To diagnose the occurrence of unknown abnormality.SOLUTION: An abnormality diagnosis device according to one aspect of the present disclosure is an abnormality diagnosis device that, based on state data representing a state of a target device and a multi-class classification model created from labeled data by supervised learning, A first abnormality diagnosis unit configured to calculate a plurality of first probabilities that the state data is classified into each of a plurality of classes including a class representing a normal state and classes respectively representing one or more known abnormalities; A second abnormality diagnosis unit configured to calculate a second probability that the state data is classified into a class indicating normality, and an integration unit configured to calculate, based on the plurality of first probabilities and the second probability, a third probability that the state data is classified into a class indicating an abnormality and one or more fourth probabilities that are classified into one or more known classes indicating one or more abnormalities.SELECTED DRAWING: Figure 2

Description

[Technical Field]

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

[0002] There are known techniques for diagnosing abnormalities in equipment such as production lines, plants, facilities, etc. For example, Patent Document 1 discloses a technique for creating an abnormality diagnosis model by supervised learning for each operating mode of a plant. Also, for example, Non-Patent Document 1 discloses a technique for creating a normal model that models the normal operating state of a plant by unsupervised learning. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7052914 [Non-patent literature]

[0004] [Non-Patent Document 1] Statistical Process Control Using Process Chemometrics, Systems / Control / Information, Vol. 48, No. 5, pp. 165-170, 2004. Summary of the Invention [Problem to be solved by the invention]

[0005] However, with conventional abnormality diagnosis techniques, when the occurrence of an abnormality is diagnosed, it is not possible to know whether the abnormality is an unknown abnormality or not.

[0006] The present disclosure has been made in consideration of the above points, and aims to make it possible to diagnose the occurrence of unknown abnormalities. [Means for solving the problem]

[0007] An abnormality diagnosis device according to one aspect of the present disclosure includes a first abnormality diagnosis unit that calculates multiple first probabilities that the status data will be classified into each of multiple classes consisting of a class representing normality and classes representing one or more known abnormalities, based on status data representing the status of a target device and a multi-class classification model created by supervised learning from labeled data; a second abnormality diagnosis unit that calculates a second probability that the status data will be classified into the class representing normality, based on the status data and a normality model created by unsupervised learning from normal data; and an integration unit that calculates a third probability that the status data will be classified into a class representing an abnormality and one or more fourth probabilities that the status data will be classified into each of the classes representing one or more known abnormalities, based on the multiple first probabilities and the second probability. [Effects of the Invention]

[0008] It is possible to diagnose the occurrence of unknown abnormalities. [Brief explanation of the drawings]

[0009] [Figure 1] FIG. 2 is a diagram illustrating an example of a hardware configuration of the abnormality diagnosis device according to the present embodiment. [Figure 2] FIG. 2 is a diagram illustrating an example of a functional configuration of the abnormality diagnosis device according to the present embodiment. [Figure 3] 10 is a flowchart illustrating an example of a model learning process according to the present embodiment. [Figure 4] 4 is a flowchart illustrating an example of an abnormality diagnosis process according to the present embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0010] An embodiment of the present invention will be described in detail below with reference to the drawings. In the following embodiment, an abnormality diagnosis device 10 will be described that can diagnose whether an unknown abnormality has occurred in equipment such as a production line, a plant, or a facility (hereinafter also referred to as "target equipment") and what the cause of the abnormality is.

[0011] Hereinafter, the data acquired from the target device will be called status data, and x = (x1, ,x N ) Each x n (n=1, . . . , N) are called state variables, and are, for example, sensor values ​​measured by a certain sensor that measures the state of the target device.

[0012] In the following, we assume that sufficient labeled state data is available, and define the set of labeled state data as D1={(x (i) ,y (i) )|i=1,...,I}. x (i) is the i-th state data, y (i) is the label for the i-th state data, and I is the total number of labeled data. (i) is the corresponding state data x (i) takes a discrete value representing the class to which y belongs, and in the following, (i) ∈{0,1,···,C}, where C is the number of known anomaly types, and y (i) =0 is the state data x (i) means that y belongs to the normal class, (i) =c∈{1, ,C} is the state data x (i) This means that the label c (0≦c≦C) belongs to the class that represents the cth known anomaly. In the following, the class represented by label c (0≦c≦C) will also be referred to as "class c".

[0013] Furthermore, in the following, we assume that sufficient status data (i.e., normal data) is obtained from the target device under normal conditions, and define the set of normal data as D2 = {x (j) |j=1, ,J}, where J is the total number of normal data.

[0014] In addition, in the following, it is assumed that sufficient time-series data of the status data acquired from the target device has been obtained, and the time-series data is defined as D3 = {x(t)|t = 1, , T}, where x(t) is the status data at discrete time t (hereinafter simply referred to as "time t"), and T is the final time of time t.

[0015] Each state data x(t) included in the time series data D3 may be labeled. i (i∈{1,2,3}) is also called a dataset. Two datasets D i and D j (i≠j, i,j∈{1,2,3}) may contain common state data.

[0016] <Example of hardware configuration of abnormality diagnosis device 10> An example of the hardware configuration of the abnormality diagnostic device 10 according to this embodiment will be described with reference to Fig. 1. Fig. 1 is a diagram showing an example of the hardware configuration of the abnormality diagnostic device 10 according to this embodiment.

[0017] 1, the abnormality diagnosis device 10 according to this embodiment includes an input device 101, a display device 102, an external I / F 103, a communication I / F 104, a RAM (Random Access Memory) 105, a ROM (Read Only Memory) 106, an auxiliary storage device 107, and a processor 108. Each of these pieces of hardware is connected to each other via a bus 109 so as to be able to communicate with each other.

[0018] The input device 101 is, for example, a keyboard, a mouse, a touch panel, a physical button, etc. The display device 102 is, for example, a display, a display panel, etc. Note that the abnormality diagnosis device 10 does not necessarily have to include at least one of the input device 101 and the display device 102, for example.

[0019] The external I / F 103 is an interface with an external device such as a recording medium 103a. Examples of the recording medium 103a include a CD (Compact Disc), a DVD (Digital Versatile Disk), an SD memory card (Secure Digital memory card), and a USB (Universal Serial Bus) memory card.

[0020] The communication I / F 104 is an interface for communicating with other devices and equipment. The RAM 105 is a volatile semiconductor memory (storage device) that temporarily stores programs and data. The ROM 106 is a non-volatile semiconductor memory (storage device) that can store programs and data even when the power is turned off. The auxiliary storage device 107 is a non-volatile storage device (storage device) such as an HDD (Hard Disk Drive), an SSD (Solid State Drive), or a flash memory. The processor 108 is one of various arithmetic devices such as a CPU (Central Processing Unit) or a GPU (Graphics Processing Unit).

[0021] 1 is an example, and the fault diagnosis device 10 may have other hardware configurations. For example, the fault diagnosis device 10 may have multiple auxiliary storage devices 107 and multiple processors 108, or may have various types of hardware other than the hardware shown in the figure.

[0022] <Example of functional configuration of abnormality diagnosis device 10> An example of the functional configuration of the abnormality diagnosis device 10 according to this embodiment will be described with reference to Fig. 2. Fig. 2 is a diagram showing an example of the functional configuration of the abnormality diagnosis device 10 according to this embodiment.

[0023] As shown in FIG. 2 , the abnormality diagnosis device 10 according to this embodiment includes a first supervised learning unit 201, a second supervised learning unit 202, an unsupervised learning unit 203, a time series analysis unit 204, a first abnormality diagnosis unit 205, a second abnormality diagnosis unit 206, a third abnormality diagnosis unit 207, a fourth abnormality diagnosis unit 208, an integration unit 209, and an output unit 210. Each of these units is implemented, for example, by a process in which one or more programs installed in the abnormality diagnosis device 10 are executed by the processor 108 or the like. The abnormality diagnosis device 10 according to this embodiment also includes a dataset storage unit 211, a first model storage unit 212, a second model storage unit 213, a third model storage unit 214, and a fourth model storage unit 215. Each of these storage units is implemented, for example, by a storage area of ​​the auxiliary storage device 107 or the like. However, at least one of the data set storage unit 211, the first model storage unit 212, the second model storage unit 213, the third model storage unit 214, and the fourth model storage unit 215 may be realized by a storage area such as a storage device (e.g., a storage device provided in a database server) that is communicatively connected to the abnormality diagnosis device 10.

[0024] The first supervised learning unit 201 uses a dataset D1 stored in a dataset storage unit 211 to create a multi-class classification model ML1 (i.e., a C+1 class classification model ML1) that classifies data into C+1 classes through supervised learning. The first supervised learning unit 201 also stores the multi-class classification model ML1 in a first model storage unit 212. Here, ML1 is a multi-class classification model that, when state data x is given, classifies the state data x into one of C+1 classes c (c=0, . . . , C). Note that any statistical or machine learning model used in supervised learning (e.g., a neural network, a support vector machine (SVM), logistic regression, etc.) can be used as the multi-class classification model ML1.

[0025] The second supervised learning unit 202 uses the dataset D1 stored in the dataset storage unit 211 to generate a two-class classification model ML2 that classifies each class c into class c and other classes through supervised learning. (c) The second supervised learning unit 202 creates a two-class classification model ML2 (c) (c=0, 1, . . . , C) is stored in the second model storage unit 213. Here, ML2 (c) is a two-class classification model that classifies state data x into class c and other classes when state data x is given. Note that each two-class classification model ML2 (c) Any statistical or machine learning model used in supervised learning (e.g., neural network, support vector machine, logistic regression, etc.) can be used as (c=0, ,C).

[0026] The unsupervised learning unit 203 creates a normal model ML3 that models the normal state of the target device through unsupervised learning using the dataset D2 stored in the dataset storage unit 211. The unsupervised learning unit 203 also stores the normal model ML3 in the third model storage unit 214. Here, ML3 is a classification model that, when given status data x, classifies the status data x as normal or other. Note that any statistical / machine learning model used in unsupervised learning (e.g., multivariate statistical process control (MSPC), one-class SVM, isolation forest, local outlier factor (LoF), etc.) can be used as the normal model ML3.

[0027] The time series analysis unit 204 uses the data set D3 stored in the data set storage unit 211 to perform time series analysis to calculate the state variable x n For each state variable x n ML4 is a time series diagnostic model that diagnoses abnormalities from time series data. (n) (n=1, . . . , N). In addition, the time series analysis unit 204 creates each time series diagnostic model ML4(n) (n=1, . . . , N) is stored in the fourth model storage unit 215. Note that each time-series diagnostic model ML4 (n) Any model (e.g., Spectral Residual (SR) method, Dynamic Time Warping (DTW), etc.) used for anomaly diagnosis using time series analysis (or may be called "anomaly detection using time series analysis") can be used as (n=1,...,N).

[0028] The first abnormality diagnosis unit 205 performs abnormality diagnosis of the target device from the state data x of the target device for abnormality diagnosis (hereinafter also referred to as "diagnosis target data x") using the multi-class classification model ML1 stored in the first model storage unit 212. That is, the first abnormality diagnosis unit 205 uses the multi-class classification model ML1 to calculate the probability P1 that the diagnosis target data x belongs to each class c. (c) (x) is calculated. Here, P1 (c) (x) is the probability that the diagnostic data x belongs to class c, and 0≦P1 (c) (x)≦1 and P1 (0) (x)+P1 (1) (x)+···+P1 (C) (x)=1. More generally, P1 (c) (x) can be said to be a score that indicates to what extent it can be said that "the diagnostic target data x belongs to class c out of classes 1 to C" (more precisely, a score that is between 0 and 1, and is normalized so that the sum of the scores for all classes is 1).

[0029] The second abnormality diagnosis unit 206 uses each two-class classification model ML2 stored in the second model storage unit 213. (c) (c=0, , C) to diagnose an abnormality of the target device from the diagnosis target data x. That is, the second abnormality diagnosis unit 206 uses the two-class classification model ML2 (c) (c=0, ,C), the probability that the diagnostic data x belongs to class c is P2 (c) (x) is calculated. Here, P2 (c)(x) is the probability that the diagnostic data x belongs to class c, and 0≦P2 (c) (x)≦1. More generally, P2 (c) (x) can be said to be a score that indicates the degree to which it can be said that "the diagnostic target data x belongs to class c" (more precisely, a score between 0 and 1, normalized so that the sum of the score when it belongs to class c and the score when it does not belong to class c is 1).

[0030] The third abnormality diagnosis unit 207 performs abnormality diagnosis of the target device from the diagnostic target data x using the normality model ML3 stored in the third model storage unit 214. That is, the third abnormality diagnosis unit 207 calculates the probability P3(x) that the diagnostic target data x is normal using the normality model ML3. Here, P3(x) is the probability that the diagnostic target data x is normal, and satisfies 0≦P3(x)≦1. More generally, P3(x) can be said to be a score that indicates to what extent it can be said that "the diagnostic target data x is normal" (more precisely, a score that is between 0 and 1, and is normalized so that the sum of the score when normal and the score when abnormal is 1).

[0031] The fourth abnormality diagnosis unit 208 calculates the time-series diagnosis model ML4 stored in the fourth model storage unit 215. (n) (n=1, . . . , N) to perform an abnormality diagnosis of the target device from time-series data including the diagnosis target data x. That is, the fourth abnormality diagnosis unit 208 uses the state variable x n For each, time series diagnostic model ML4 (n) Using the state variable x n This calculates information indicating whether the time series data for the state variable x is normal or abnormal. n (n=1, ,N) which state variable x n is abnormal, or in other words, the state variable x that is the cause of the abnormality n is obtained.

[0032] The integrating unit 209 calculates an abnormality diagnosis result by integrating the abnormality diagnosis result by the first abnormality diagnosis unit 205, the abnormality diagnosis result by the second abnormality diagnosis unit 206, and the abnormality diagnosis result by the third abnormality diagnosis unit 207. In this way, the abnormality diagnosis result by supervised learning and the abnormality diagnosis result by unsupervised learning are integrated, making it possible to estimate the occurrence of an unknown abnormality.

[0033] The output unit 210 outputs the abnormality diagnosis result integrated by the integration unit 209 and the abnormality diagnosis result by the fourth abnormality diagnosis unit 208 to a predetermined output destination.

[0034] The data set storage unit 211 stores a data set D1, a data set D2, and a data set D3.

[0035] The first model storage unit 212 stores the multi-class classification model ML1 created by the first supervised learning unit 201.

[0036] The second model storage unit 213 stores the two-class classification model ML2 generated by the second supervised learning unit 202. (c) Store (c=0,1,···,C).

[0037] The third model storage unit 214 stores the normal model ML3 created by the unsupervised learning unit 203.

[0038] The fourth model storage unit 215 stores each time series diagnostic model ML4 created by the time series analysis unit 204. (n) (n=1, ,N) is stored.

[0039] <Model learning process> An example of the model learning process according to this embodiment will be described with reference to Fig. 3. Fig. 3 is a flowchart showing an example of the model learning process according to this embodiment. Note that steps S101 to S104 in Fig. 3 are in no particular order and may be executed in any order.

[0040] The first supervised learning unit 201 creates a multi-class classification model ML1 through supervised learning using a dataset D1 stored in the dataset storage unit 211 (step S101). The multi-class classification model ML1 is stored in the first model storage unit 212.

[0041] The second supervised learning unit 202 uses the dataset D1 stored in the dataset storage unit 211 to generate a two-class classification model ML2 for each class c through supervised learning. (c) (c=0, 1, , C) is created (step S102). (c) (c=0, 1, . . . , C) is stored in the second model storage unit 213.

[0042] The unsupervised learning unit 203 creates a normal model ML3 by unsupervised learning using the dataset D2 stored in the dataset storage unit 211 (step S103). The normal model ML3 is stored in the third model storage unit 214.

[0043] The time series analysis unit 204 uses the data set D3 stored in the data set storage unit 211 to perform time series analysis to calculate the state variable x n For each, time series diagnostic model ML4 (n) (n=1, , N) is created (step S104). (n) (n=1, . . . , N) is stored in the fourth model storage unit 215.

[0044] <Abnormality diagnosis processing> An example of the abnormality diagnosis process according to this embodiment will be described with reference to FIG. 4. FIG. 4 is a flowchart showing an example of the abnormality diagnosis process according to this embodiment. Hereinafter, the abnormality diagnosis process will be described with reference to the diagnosis object data x and the diagnosis object data x=x(t p ) containing time series data X={x(t)|t=t p ,t p -1,...,t p -Δt} is given, where t pis the time when the diagnostic object data x is acquired (for example, the current time), and Δt is the length of the time-series data X. Note that steps S201 to S203 in Fig. 4 are in no particular order and may be executed in any order. Furthermore, steps S201 to S204 and step S205 may be executed in parallel or sequentially.

[0045] The first abnormality diagnosis unit 205 performs abnormality diagnosis of the target device from the diagnostic object data x using the multi-class classification model ML1 stored in the first model storage unit 212 (step S201). As a result, the probability P1 (c) (x)(c=0,1,···,C) is obtained.

[0046] The second abnormality diagnosis unit 206 uses each two-class classification model ML2 stored in the second model storage unit 213. (c) Using (c=0, , C), an abnormality diagnosis of the target device is performed from the diagnostic target data x (step S202). As a result, for each class c∈{0, 1, , C}, the probability P2 (c) (x) is obtained.

[0047] The third abnormality diagnosis unit 207 performs abnormality diagnosis of the target device from the diagnostic target data x using the normal model ML3 stored in the third model storage unit 214 (step S203). As a result, the probability P3(x) that the diagnostic target data x is normal is obtained.

[0048] The integrating unit 209 integrates the abnormality diagnosis results from the above steps S201 to S203 (step S204). The integrating unit 209 integrates the abnormality diagnosis results, for example, in the following manner.

[0049] P (OK) (x)=(P1 (0) (x)+P2 (0) (x)+P3(x)) / 3 P (NG) (x)=1-P (OK) (x) P(c) (x)=(P1 (c) (x)+P2 (c) (x)) / 2 (c∈{1, ,C}) where P (OK) (x) is the probability that the diagnostic data x belongs to the normal class, P (NG) (x) is the probability that the diagnostic target data x belongs to the class that indicates an abnormality. (c) (x) is the probability that the diagnostic target data x belongs to class c (i.e., the probability that the diagnostic target data x belongs to class c, which represents the cth known abnormality). (NG) While the value of (x) is high, all P (c) If the value of (x) is low, it can be estimated that there is a high possibility that an unknown abnormality has occurred.

[0050] The fourth abnormality diagnosis unit 208 calculates the state variable x n For each (n=1, . . . , N), each time-series diagnostic model ML4 stored in the fourth model storage unit 215 (n) The fourth abnormality diagnosis unit 208 performs abnormality diagnosis of the target device from the time-series data X including the diagnosis target data x using the state variable x n For each state variable x n ML4 time series diagnostic model corresponding to (n) Using the state variable x n Time series data for X n ={x n (t)|t=t p ,t p -1,...,t p -Δt} is normal or abnormal. n For each time series data X n Information r indicating whether the n ∈{0,1}, where r n =0 is time series data X n means that it is normal, and r n =1 is time series data X n This means that if the target device is abnormal,n = 1, and the state variable x n At this time, the fourth abnormality diagnosis unit 208 determines whether the time series data X n Information r indicating whether the n A score d that represents the confidence level of n may be calculated.

[0051] The output unit 210 outputs the abnormality diagnosis result integrated in the above step S204 and the abnormality diagnosis result in the above step S205 to a predetermined output destination (step S206). (OK) (x) and P (NG) (x) and P (c) (x)(c∈{1, ,C}) and (r n ,d n ) (n∈{1, ,N}) to the output destination. Note that the output destination is not limited to a specific output destination and can be any output destination, and examples thereof include the display device 102 such as a display, a storage area of ​​the auxiliary storage device 107, and a device (e.g., a control device that controls the target device) that is communicably connected to the abnormality diagnosis device 10.

[0052] <Modification> A modification of the above embodiment will now be described.

[0053] Variation 1 The time series analysis unit 204, the fourth abnormality diagnosis unit 208, and the fourth model storage unit 215 are not essential components, and the abnormality diagnosis device 10 does not have to have these units. In this case, it is not possible to estimate the state variable that is the cause of the abnormality, but it is possible to estimate the occurrence of an unknown abnormality.

[0054] Variation 2 The second supervised learning unit 202, the second abnormality diagnosis unit 206, and the second model storage unit 213 are not essential components, and the abnormality diagnosis device 10 does not necessarily have to include these units. In this case, the integrating unit 209 may integrate the abnormality diagnosis results, for example, as follows.

[0055] P (OK) (x)=(P1 (0) (x)+P3(x)) / 2 P (NG) (x)=1-P (OK) (x) P (c) (x)=P1 (c) (x) (c∈{1, ,C}) In this case, although the accuracy of the abnormality diagnosis results after integration decreases, it becomes possible to perform abnormality diagnosis that is lighter and memory-saving and that can estimate the occurrence of unknown abnormalities.

[0056] Variation 3 When the integration unit 209 integrates the respective abnormality diagnosis results, weighted integration may be performed. For example, by using α1, α2, and α3 that satisfy α1+α2+α3=1, P (OK) (x)=α1P1 (0) (x)+α2P2 (0) (x) + α3P3(x). Similarly, for example, by using β1 and β2 that satisfy β1 + β2 = 1, P (c) (x)=β1P1 (c) (x)+β2P2 (c) May be integrated with (x).

[0057] Variation 4 The integration unit 209 calculates, for example, Q(x)=1−(P (1) (x)+···+P (C) (x)) / C may be calculated.

[0058] Variation 5 Two or more of the above-described modifications 1 to 4 may be combined.

[0059] <Summary> As described above, the abnormality diagnosis device 10 according to this embodiment can estimate whether an abnormality occurring in a target device is an unknown abnormality by using at least a multi-class classification model created by supervised learning and a normal model created by unsupervised learning. In addition, the abnormality diagnosis device 10 according to this embodiment can also estimate the state variables that are causing the abnormality occurring in the target device. Therefore, for example, even if an unknown abnormality occurs in the target device, a person in charge, such as an operator of the target device, can take early action against the abnormality.

[0060] The present invention is not limited to the above-described specifically disclosed embodiments, and various modifications, changes, and combinations with known technologies are possible without departing from the scope of the claims. [Explanation of symbols]

[0061] 10. Abnormality diagnosis device 101 Input Device 102 Display device 103 External I / F 103a Recording media 104 Communication I / F 105 RAM 106 ROM 107 Auxiliary storage 108 processors 109 Bus 201 First Supervised Learning Unit 202 Second Supervised Learning Unit 203 Unsupervised Learning 204 Time Series Analysis Section 205 First abnormality diagnosis section 206 Second abnormality diagnosis section 207 Third abnormality diagnosis section 208 Fourth Abnormality Diagnosis Unit 209 Integration Department 210 Output section 211 Dataset Storage Unit 212 First model storage unit 213 Second Model Storage Unit 214 Third Model Memory Unit 215 Fourth Model Memory Unit

Claims

1. a first abnormality diagnosis unit that calculates a plurality of first probabilities that the status data will be classified into each of a plurality of classes consisting of a class representing normality and classes representing one or more known abnormalities, based on status data representing a status of the target device and a multi-class classification model created by supervised learning from labeled data; and a second abnormality diagnosis unit that calculates a second probability that the status data will be classified into a class representing normality based on the status data and a normality model created from normal data by unsupervised learning; an integration unit that calculates a third probability that the status data will be classified into a class representing an abnormality and one or more fourth probabilities that the status data will be classified into each of one or more classes representing known abnormalities, based on the plurality of first probabilities and the second probabilities; An abnormality diagnosis device having the above structure.

2. a third abnormality diagnosis unit that calculates, for each of a plurality of classes including a class representing normality and classes representing one or more known abnormalities, a fifth probability that the condition data will be classified into the class and a class other than the class, based on the condition data and a plurality of two-class classification models created by supervised learning from labeled data; The integration unit The abnormality diagnosis device according to claim 1 , wherein the third probability and the one or more fourth probabilities are calculated based on the fifth probability as well.

3. The integration unit 3. The abnormality diagnosis device according to claim 1, further comprising: a sixth probability that the status data is classified into a class representing an unknown abnormality.

4. The integration unit calculating an average value of a first probability that the status data is classified into the class representing normal, the second probability, and a fifth probability that the status data is classified into the class representing normal, and calculating a value obtained by subtracting the average value from 1 as the third probability; 3. The abnormality diagnosis device according to claim 2, wherein, for each class representing a known abnormality, the fourth probability is calculated as an average value of a first probability that the status data is classified into the class and a fifth probability that the status data is classified into the class.

5. 2. The abnormality diagnosis device according to claim 1, further comprising: a fourth abnormality diagnosis unit that diagnoses whether or not an abnormality has occurred for each of the state variables based on the time series data of the state data and a time series diagnosis model created by time series analysis from the time series data of the state variables for each of the state variables constituting the state data.

6. a first abnormality diagnosis procedure for calculating a plurality of first probabilities that the status data will be classified into each of a plurality of classes consisting of a class representing normality and classes representing one or more known abnormalities, based on status data representing the status of the target device and a multi-class classification model created by supervised learning from labeled data; a second abnormality diagnosis procedure for calculating a second probability that the status data will be classified into a class representing normality based on the status data and a normality model created from normal data by unsupervised learning; an integration procedure for calculating a third probability that the status data will be classified into a class representing an abnormality and one or more fourth probabilities that the status data will be classified into each of one or more classes representing known abnormalities, based on the plurality of first probabilities and the second probabilities; The computer performs an abnormality diagnosis method.

7. a first abnormality diagnosis procedure for calculating a plurality of first probabilities that the status data will be classified into each of a plurality of classes consisting of a class representing normality and classes representing one or more known abnormalities, based on status data representing the status of the target device and a multi-class classification model created by supervised learning from labeled data; a second abnormality diagnosis procedure for calculating a second probability that the status data will be classified into a class representing normality based on the status data and a normality model created from normal data by unsupervised learning; an integration procedure for calculating a third probability that the status data will be classified into a class representing an abnormality and one or more fourth probabilities that the status data will be classified into each of one or more classes representing known abnormalities, based on the plurality of first probabilities and the second probabilities; A program that causes a computer to execute the following.

Images