JP7371802B1 - Abnormality diagnosis system, abnormality diagnosis device, abnormality diagnosis method, and program - Google Patents

Abstract

The present invention provides a technology that can reduce the management cost of an abnormality diagnosis model. An abnormality diagnosis system performs abnormality diagnosis of equipment that includes a plurality of abnormality diagnosis targets having a plurality of operation modes. The first acquisition part of the first data set, which is learning data representing past operation data of the equipment, and the learning data included in the first data set are classified into groups according to the operation mode targeted for abnormality diagnosis. a classification unit that creates a second data set; a first creation unit that creates learning setting information for learning an abnormality diagnosis model for each operation mode of each abnormality diagnosis target based on the second data set; A learning unit that learns an abnormality diagnosis model for an abnormality diagnosis target and operation mode using state variable values of a part of learning data classified into groups in a second data set for each group; and a second creation unit that creates model management information for managing the abnormality diagnosis model for each operation mode based on the abnormality diagnosis model. [Selection diagram] Figure 2

Description

The present disclosure relates to an abnormality diagnosis system, an abnormality diagnosis apparatus, an abnormality diagnosis method, and a program.

Generally, measuring instruments are attached to a plant, and the plant is controlled by a control device based on values measured by these measuring instruments. Since the data acquired from these measuring instruments and control devices represents the operating status of the plant, it is also called operating data.

A known technique is to construct an abnormality diagnosis model using plant operation data and then perform abnormality diagnosis using the abnormality diagnosis model. For example, Patent Document 1 discloses a technique in which abnormality diagnosis models are learned for each operating mode of a plant and then abnormality diagnosis is performed using these abnormality diagnosis models.

Patent No. 7052914

However, conventionally, when multiple abnormality diagnosis models are used for abnormality diagnosis (including the case where one abnormality diagnosis model selected from multiple abnormality diagnosis models is used for abnormality diagnosis), each abnormality diagnosis Models had to be trained and managed separately. Therefore, when the number of abnormality diagnosis models increases, the cost of managing the abnormality diagnosis models increases.

The present disclosure has been made in view of the above points, and provides a technique that can reduce the management cost of an abnormality diagnosis model.

An abnormality diagnosis system according to one aspect of the present disclosure is an abnormality diagnosis system that performs abnormality diagnosis of equipment that is configured of a plurality of abnormality diagnosis targets in which one or more operation modes exist, and that uses past operation data of the equipment. a first acquisition unit configured to acquire a first data set constituted by learning data representing the learning data; a classification unit configured to create a second data set classified into groups; and for learning an abnormality diagnosis model for each operation mode of each abnormality diagnosis target based at least on the second data set. a first creation unit configured to create learning setting information for each group; and a first creation unit configured to create learning setting information for each group, of learning data classified into the group from among the learning data included in the second data set. a learning unit configured to learn an abnormality diagnosis model for an abnormality diagnosis target and operation mode corresponding to the group using at least some of the state variable values, the learning setting information, and the abnormality diagnosis model; a second creation unit configured to create model management information for managing an abnormality diagnosis model for each operation mode of each abnormality diagnosis target based on the above.

A technique is provided that can reduce the management cost of an abnormality diagnosis model.

It is a figure showing an example of an operation mode. 1 is a diagram showing an example of the overall configuration of an abnormality diagnosis system according to the present embodiment. FIG. 1 is a diagram showing an example of the hardware configuration of a model learning management device according to the present embodiment. 1 is a diagram illustrating an example of a hardware configuration of an abnormality diagnosis device according to an embodiment. FIG. 3 is a diagram illustrating an example of the functional configuration of a model learning management processing unit according to the present embodiment. FIG. 3 is a diagram showing an example of a functional configuration of an abnormality diagnosis processing section according to the present embodiment. 7 is a flowchart illustrating an example of model learning management processing according to the present embodiment. FIG. 3 is a diagram showing an example of a learning data set. FIG. 3 is a diagram showing an example of a sub-learning data set. It is a figure which shows an example of a learning setting information table. It is a figure which shows an example of a model number information table. It is a figure which shows an example of a model detailed information table. 7 is a flowchart illustrating an example of an abnormality diagnosis process according to the present embodiment.

An embodiment of the present invention will be described below. In the following embodiments, a case where a plurality of abnormality diagnosis models are used for abnormality diagnosis (including a case where one abnormality diagnosis model selected from a plurality of abnormality diagnosis models is used for abnormality diagnosis) is targeted. An abnormality diagnosis system 1 that can reduce the management cost of abnormality diagnosis models will be described.

Here, production facilities such as plants are generally composed of a plurality of devices, devices, facilities, etc. (hereinafter, devices, devices, facilities, etc. are collectively referred to simply as "equipment"). For this reason, when an abnormality diagnosis is performed on production equipment such as a plant, multiple abnormality diagnoses are performed to perform abnormality diagnosis on each of the multiple equipment that is the target of abnormality diagnosis among the multiple equipment that constitutes the production equipment. Models are commonly used.

Further, equipment constituting production equipment such as a plant may have a plurality of operation modes (or may be referred to as operation modes). For this reason, for devices having multiple operation modes, it is common to use multiple abnormality diagnosis models for diagnosing abnormalities in each of the multiple operation modes.

<Operating mode>
An operating mode is an operating method that indicates how a device is operated. For example, assuming a steam boiler as the equipment, the operating modes include "high pressure mode," which is the operating mode when the pressure inside the steam boiler is high, and "pressure high mode," which is the operating mode when the pressure inside the steam boiler is low. "low mode" etc. For example, if we assume gas piping equipment as the equipment, the operating mode may be "gas temperature high mode", which is the operating mode when the gas temperature in the piping is high, or "gas temperature high mode", which is the operation mode when the gas temperature in the piping is low. Examples include "gas temperature low mode" which is an operation mode. However, these are just examples, and various operation modes may exist depending on the type and type of equipment. Further, it is assumed that it is known what kind of operation mode exists in the device that is the target of abnormality diagnosis.

In general, operational data for production equipment such as plants includes the values of various state variables such as temperature, pressure, and flow rate measured by measuring instruments. It characteristically appears in the values of a specific state variable (hereinafter also referred to as the main variable) and variables related to it (hereinafter also referred to as the related variable). Therefore, the operation mode can be specified from the main variable and related variables. Note that there may be multiple related variables for one main variable.

For example, FIG. 1 shows an operating mode during normal operation of a certain device in which the main variable is "x ₁ " and the related variable is "x _2. " In the example shown in FIG. 1, under normal operating conditions, in operation mode 1, the value of main variable x ₁ is less than a certain predetermined threshold, and there is no correlation between main variable x ₁ and related variable x ₂ , and in operation mode 2 In operation mode 3, there is a positive correlation between the main variable x ₁ and the related variable x ₂ , and in operation mode 3, the value of the main variable x ₁ is above a certain predetermined threshold and there is no correlation between the main variable x ₁ and the related variable x ₂ . This characteristic is expressed. By using such characteristics, when operational data is given, it is possible to classify this operational data into one of the operating modes. Note that in the following, if a device to be subjected to abnormality diagnosis does not have a plurality of operation modes, it will be assumed that the device has only one operation mode.

<Example of overall configuration of abnormality diagnosis system 1>
FIG. 2 shows an example of the overall configuration of the abnormality diagnosis system 1 according to this embodiment. As shown in FIG. 2, the abnormality diagnosis system 1 according to the present embodiment includes a model learning management device 10, an abnormality diagnosis device 20, a production facility 30, a measuring instrument 40, and a control device 50. Furthermore, the model learning management device 10 and the abnormality diagnosis device 20 are communicably connected via a communication network. Similarly, the abnormality diagnosis device 20 and the measuring instrument 40 are communicably connected via a communication network. Similarly, the abnormality diagnosis device 20 and the control device 50 are communicably connected via a communication network.

The model learning management device 10 is capable of creating abnormality diagnosis models used for abnormality diagnosis of each abnormality diagnosis target 310 constituting the production equipment 30 and creating information for managing these abnormality diagnosis models. A computer or computer system that can As the model learning management device 10, it is possible to use, for example, a PC (personal computer), a general-purpose server, or the like.

The abnormality diagnosis device 20 is a computer or computer system that can perform abnormality diagnosis on each abnormality diagnosis target 310 that constitutes the production equipment 30. As the abnormality diagnosis device 20, it is possible to use, for example, a PC, a general-purpose server, or the like.

The production facility 30 is, for example, a plant configured with a plurality of devices. The production facility 30 includes a plurality of abnormality diagnosis targets 310, which are a plurality of devices to be subjected to abnormality diagnosis. These plurality of abnormality diagnosis targets 310 are some of the plurality of devices constituting the production facility 30 . Furthermore, one or more measuring instruments 40 for measuring the values of state variables of the production equipment 30 are installed in some of the plurality of equipment that constitute the production equipment 30 . Note that specific examples of the production equipment 30 include, for example, energy plants, petrochemical manufacturing plants, food and pharmaceutical manufacturing plants, and the like.

Hereinafter, where the number of state variables of the production equipment 30 is n, data composed of the values of n state variables x ₁ , . . . , x _n will be referred to as operation data, and x = (x ₁ , . . . , x _n ). For example, since the driving data is obtained every measurement period of the measuring instrument 40, when specifying the time t of the driving data, x(t)=(x ₁ (t),..., x _n (t)) It is expressed as Note that the time t may be the acquisition time of the driving data or the measurement time of the state variable value included in the driving data.

The control device 50 is a computer or computer system that controls the operation of the production equipment 30 using the operation data x acquired from the measuring instrument 40. As the control device 50, it is possible to use, for example, a PLC (Programmable Logic Controller), a DCS (Distributed Control System), or the like.

Here, the model learning management device 10 includes a model learning management processing section 110. The model learning management processing unit 110 uses past driving data stored in the driving database 220 as learning data to learn an abnormality diagnosis model for diagnosing each abnormality diagnosis target 310. Furthermore, the model learning management processing unit 110 creates information for managing each abnormality diagnosis model (hereinafter also referred to as model management information).

Further, the abnormality diagnosis device 20 includes an abnormality diagnosis processing section 210 and an operation database 220. The abnormality diagnosis processing unit 210 performs abnormality diagnosis on the abnormality diagnosis target 310 using new operation data (hereinafter also referred to as diagnostic data) stored in the operation database 220, model management information, and an abnormality diagnosis model. I do. The operation database 220 stores operation data obtained from the measuring instrument 40, the control device 50, or both.

Note that the overall configuration of the abnormality diagnosis system 1 shown in FIG. 2 is an example, and is not limited to this. For example, the model learning management device 10 and the abnormality diagnosis device 20 may be configured integrally, or, for example, a device having the operation database 220 may exist separately from the abnormality diagnosis device 20. Further, for example, the abnormality diagnosis system 1 may include a terminal device used by an operator, a manager, or the like of the production equipment 30.

<Example of hardware configuration of model learning management device 10>
FIG. 3 shows an example of the hardware configuration of the model learning management device 10 according to this embodiment. As shown in FIG. 3, the model learning management device 10 according to the present embodiment includes an input device 401, a display device 402, an external I/F 403, a communication I/F 404, and a RAM (Random Access Memory) 405. It has a ROM (Read Only Memory) 406, an auxiliary storage device 407, and a processor 408. Each of these pieces of hardware is communicably connected via a bus 409.

The input device 401 is, for example, a keyboard, a mouse, a touch panel, a physical button, or the like. The display device 402 is, for example, a display, a display panel, or the like. Note that the model learning management device 10 may not include at least one of the input device 401 and the display device 402, for example.

The external I/F 403 is an interface with an external device such as a recording medium 403a. Examples of the recording medium 403a 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.

Communication I/F 404 is an interface for connecting model learning management device 10 to a communication network. The RAM 405 is a volatile semiconductor memory (storage device) that temporarily holds programs and data. The ROM 406 is a nonvolatile semiconductor memory (storage device) that can retain programs and data even when the power is turned off. The auxiliary storage device 407 is, for example, a nonvolatile storage device such as an HDD (Hard Disk Drive), an SSD (Solid State Drive), or a flash memory. The processor 408 is, for example, various arithmetic devices such as a CPU (Central Processing Unit). Here, the model learning management processing unit 110 included in the model learning management device 10 is realized, for example, by a process that one or more programs installed in the model learning management device 10 causes the processor 408 or the like to execute.

Note that the hardware configuration of the model learning management device 10 shown in FIG. 3 is an example, and is not limited to this. For example, the model learning management device 10 may have a plurality of auxiliary storage devices 407 and a plurality of processors 408, may not have a part of the illustrated hardware, or may not have a part of the illustrated hardware. It may also include various hardware other than the above.

<Example of hardware configuration of abnormality diagnosis device 20>
FIG. 4 shows an example of the hardware configuration of the abnormality diagnosis device 20 according to this embodiment. As shown in FIG. 4, the abnormality diagnosis device 20 according to the present embodiment includes an input device 501, a display device 502, an external I/F 503, a communication I/F 504, a RAM 505, a ROM 506, and an auxiliary storage device 507. and a processor 508. Each of these pieces of hardware is communicably connected via a bus 509.

The input device 501 is, for example, a keyboard, a mouse, a touch panel, a physical button, or the like. The display device 502 is, for example, a display, a display panel, or the like. Note that the abnormality diagnosis device 20 does not need to have at least one of the input device 501 and the display device 502.

The external I/F 503 is an interface with an external device such as a recording medium 503a. Examples of the recording medium 503a include a CD, DVD, SD memory card, and USB memory card.

Communication I/F 504 is an interface for connecting abnormality diagnosis device 20 to a communication network. The RAM 505 is a volatile semiconductor memory (storage device) that temporarily holds programs and data. The ROM 506 is a nonvolatile semiconductor memory (storage device) that can retain programs and data even when the power is turned off. The auxiliary storage device 507 is, for example, a nonvolatile storage device such as an HDD, SSD, or flash memory. The processor 508 is, for example, various arithmetic devices such as a CPU. Here, the abnormality diagnosis processing unit 210 included in the abnormality diagnosis apparatus 20 is realized by, for example, a process that one or more programs installed in the abnormality diagnosis apparatus 20 causes the processor 508 or the like to execute. Further, the operation database 220 included in the abnormality diagnosis device 20 is realized by, for example, a storage area of the auxiliary storage device 507.

Note that the hardware configuration of the abnormality diagnosis device 20 shown in FIG. 4 is an example, and is not limited to this. For example, the abnormality diagnosis device 20 may include multiple auxiliary storage devices 507 and multiple processors 508, may not include some of the illustrated hardware, or may include hardware other than the illustrated hardware. may include various types of hardware.

<Detailed functional configuration example of model learning management processing unit 110>
FIG. 5 shows a detailed functional configuration example of the model learning management processing unit 110 according to this embodiment. As shown in FIG. 5, the model learning management processing unit 110 according to the present embodiment includes a learning data acquisition unit 111, a learning data classification unit 112, an input/output unit 113, and a learning setting information creation unit 114. , and a learning section 115.

The learning data acquisition unit 111 acquires each of a plurality of past driving data stored in the driving database 220 as learning data. Hereinafter, a set of a plurality of learning data acquired by the learning data acquisition unit 111 will also be referred to as a learning data set. Further, in the following, it is assumed that each learning data included in the learning data set is given a sample number as identification information for identifying the learning data within the learning data set.

The learning data classification unit 112 classifies the learning data included in the learning data set into groups for each abnormality diagnosis target 310. At this time, one piece of learning data may be classified into two or more groups.

In addition, for each group, the learning data classification unit 112 includes the sample number of the learning data belonging to the group, the main variable of the abnormality diagnosis target 310 corresponding to the group, and the state variable value of the learning data. Data (hereinafter also referred to as sub-learning data) consisting of values of related variables is created, and these sub-learning data are further classified into groups for each operation mode. Hereinafter, a set of these sub-learning data will also be referred to as a sub-learning data set, and a set of sub-learning data classified into one of the plurality of groups will also be referred to as a group-based sub-learning data set. The sub-learning data set is a union of sub-learning data sets for each group.

In addition, when further classifying the sub-learning data into groups for each operation mode, the learning data classification unit 112 may perform the classification using a method similar to the method described in Patent Document 1 mentioned above.

The input/output unit 113 accepts user input from the input device 401 and causes the display device 402 to display various screens. For example, the input/output unit 113 displays a screen that visualizes the learning data set and the sub-learning data set on the display device 402 such as a display, and also performs an operation for creating a learning setting information table (hereinafter referred to as learning data set). (also referred to as a setting information table creation operation) is received from the input device 401. Here, the learning setting information table is information indicating one or more group-based sub-learning data sets for learning one or more abnormality diagnosis models used for abnormality diagnosis of the abnormality diagnosis target 310 (hereinafter referred to as group-based (Also referred to as a sub-learning data set number.) This is data in a table format in which learning setting information is stored. Note that a specific example of the learning setting information table will be described later.

The learning setting information creation unit 114 creates a learning setting information table. For example, the learning setting information creation unit 114 creates a learning setting information table based on the learning setting information table creation operation accepted by the input/output unit 113. However, the present invention is not limited to this, and the learning setting information creation unit 114 may create the learning setting information table using, for example, the sub-learning data set or both the learning data set and the sub-learning data set.

The learning unit 115 creates an abnormality diagnosis model using the learning setting information table and the sub-learning data set. For example, the learning unit 115 uses, for each piece of learning setting information stored in the learning setting information table, the group-based sub-learning data set indicated by the group-based sub-learning data set number included in the learning setting information. An abnormality diagnosis model used for abnormality diagnosis of the abnormality diagnosis target 310 corresponding to the group-based sub-learning data set number is learned.

Furthermore, the learning unit 115 creates model management information that is information for managing each abnormality diagnosis model. Here, the model management information includes a model number information table and a model detailed information table. The model number information table is data in a table format in which model number information representing the number of abnormality diagnosis models of the abnormality diagnosis target 310 is stored. The model detailed information table is the detailed information of the abnormality diagnosis model (for example, the name of the abnormality diagnosis target 310 for which abnormality diagnosis is performed using the abnormality diagnosis model, the conditions under which the abnormality diagnosis model is used, the abnormality diagnosis This is data in a table format that stores detailed model information representing state variables (also called explanatory variables) that the model takes as input. Note that specific examples of the model number information table and model detailed information table will be described later.

Further, the learning unit 115 transmits each of the above-mentioned abnormality diagnosis models and model management information to the abnormality diagnosis processing unit 210 of the abnormality diagnosis device 20. However, this is just an example, and for example, the learning unit 115 may write each of the above abnormality diagnosis models and model management information to the recording medium 403a or the like. In this case, the abnormality diagnosis processing unit 210 of the abnormality diagnosis device 20 may read each abnormality diagnosis model and model management information from the recording medium 403a or the like.

<Detailed functional configuration example of abnormality diagnosis processing unit 210>
FIG. 6 shows a detailed functional configuration example of the abnormality diagnosis processing section 210 according to this embodiment. As shown in FIG. 6, the abnormality diagnosis processing section 210 according to this embodiment includes a diagnostic data acquisition section 211, an abnormality diagnosis section 212, and an input/output section 213.

The diagnostic data acquisition unit 211 acquires new driving data stored in the driving database 220 as diagnostic data.

The abnormality diagnosis unit 212 refers to the model management information to identify an abnormality diagnosis model to be used for abnormality diagnosis of each abnormality diagnosis target 310, and then performs abnormality diagnosis of each abnormality diagnosis target 310 using the specified abnormality diagnosis model. conduct.

The input/output unit 213 accepts user input from the input device 501 and causes the display device 502 to display various screens. For example, the input/output unit 213 displays a screen that visualizes the abnormality diagnosis results of each abnormality diagnosis target 310 on the display device 502 such as a display.

<Model learning management processing>
The model learning management process according to this embodiment will be described below with reference to FIG. 7. It is assumed that the operation database 220 stores past operation data of the production equipment 30.

First, the learning data acquisition unit 111 of the model learning management processing unit 110 acquires each of a plurality of past driving data stored in the driving database 220 as learning data (step S101). As a result, a learning data set composed of these learning data is obtained. For example, if the training data set is D, the sample number is i, and the training data with sample number i is x ⁽ⁱ⁾ , then the training data set is D = {(i, x ⁽ⁱ⁾ ) It is expressed as |i=1,...,m}. Here, m is the number of learning data included in the learning data set D, and is, for example, a predetermined integer of 1 or more. A specific example of the learning data set is shown in FIG. The learning data set shown in FIG. 8 includes the following learning data x ⁽¹⁾ , x ⁽²⁾ , x ⁽³⁾ , ..., x ^(m) .

x ⁽¹⁾ = (1013, 0.54, 815.0,..., 78.2)
x ⁽²⁾ = (1015, 0.58, 822.3,..., 79.5)
x ⁽³⁾ = (1012, 0.68, 798.6,..., 102.3)
...
x ^(m) = (1013, 0.52, 818.1,..., 95.4)
Here, for example, when learning an abnormality diagnosis model using a multivariate statistical process control (MSPC) method, only past operating data of the production equipment 30 during normal operation is acquired as learning data. . For details on multivariate statistical process management, please refer to Reference 1, for example.

On the other hand, for example, when learning an abnormality diagnosis model using a machine learning method, not only past operation data of the production equipment 30 during normal times but also past operation data of the production equipment 30 during abnormal times is used as learning data. You may obtain it. However, in this case, the learning data is given a label y indicating whether the data is operating data when the production equipment 30 is in a normal state or in an abnormal state. That is, in this case, the learning data set D is expressed as D={(i, x ⁽ⁱ⁾ , y ⁽ⁱ⁾ ) | i=1, . . . , m}. y ⁽ⁱ⁾ is a label of the learning data x ⁽ⁱ⁾ , and takes, for example, a binary value indicating whether it is an abnormal time or a normal time.

Next, the learning data classification section 112 of the model learning management processing section 110 creates a sub-learning dataset from the learning dataset D (step S102). The learning data classification unit 112 may create a sub-learning data set by, for example, steps 11 and 12 below.

Step 11: The learning data classification unit 112 uses schedule information including the operation start time and operation end time of each abnormality diagnosis target 310 constituting the production equipment 30 to classify the learning data while the abnormality diagnosis target 310 is in operation. The data is classified into a group (hereinafter also referred to as a first group) corresponding to the abnormality diagnosis target 310. That is, for example, when the operation start time of a certain abnormality diagnosis target 310 is t _s and the operation end time is _te , the learning data classification unit 112 classifies the operation start time t s and the operation end time te from the operation start time t _s to the operation end time _te . The learning data is classified into a first group corresponding to the abnormality diagnosis target 310. Thereby, the learning data included in the learning data set D is classified into the first group for each abnormality diagnosis target 310. It should be noted that since a plurality of abnormality diagnosis targets 310 may be operating at the same time, one learning data may be classified into a plurality of first groups.

Step 12: The learning data classification unit 112 creates sub-learning data for each first group from the learning data classified into the first group, and assigns these sub-learning data to the corresponding learning data. The abnormality diagnosis targets 310 corresponding to the first group are further classified into groups for each operation mode (hereinafter also referred to as a second group). Note that the abnormality diagnosis target 310 may have one operation mode, and in this case, the sub-learning data created from the learning data classified into a certain first group are divided into one second group. will be classified into the following groups.

Here, the learning data classification unit 112 selects the main variable and related variables of the abnormality diagnosis target 310, and then selects the sample number of the learning data belonging to the first group corresponding to the abnormality diagnosis target 310, Data composed of the main variable and the values of related variables of the learning data may be created as sub-learning data. At this time, the learning data classification unit 112 may select the main variable and the related variables using a method similar to the method described in Patent Document 1 mentioned above. Note that the above-mentioned Patent Document 1 discloses, as methods for selecting the main variables, a method in which a state variable selected by the user is selected as the main variable, and a method in which the main variable is selected based on business knowledge regarding the abnormality diagnosis target 310, etc. is disclosed. Further, in the above-mentioned Patent Document 1, as related variable selection methods, a method of selecting a state variable selected by a user as a related variable, and a method of selecting a selected variable based on business knowledge regarding the abnormality diagnosis target 310, etc. Then, the degree of association between the main variable and state variables other than the main variable (for example, correlation coefficient, MIC (Maximal Information Coefficient), HSIC (Hilbert-Schmidt Independence Criteria), etc.) is calculated, and the predetermined A method for selecting a number of state variables as relevant variables is disclosed.

Further, the learning data classification unit 112 classifies the sub-learning data for each operation mode of the abnormality diagnosis target 310 corresponding to the first group using a method similar to the method described in Patent Document 1 mentioned above. They may be further classified into the second group. In addition, in the above-mentioned Patent Document 1, as a method of classifying sub-learning data into a second group for each operation mode, a user's classification operation (for example, There are two methods: classification based on point cloud range specification operations, range specification operations for main variables and related variables, etc.), classification based on the start and end times of each operation mode, and similarity between sub-learning data. A method of calculating a degree of similarity (for example, cosine similarity, etc.) and classifying sub-learning data whose similarity is within a predetermined range as the same group, and a method of classifying by clustering are disclosed.

Through steps 11 and 12 above, sub-learning data sets classified into a plurality of second groups are obtained. Hereinafter, the sub-learning dataset will be referred to as S, and the group-based sub-learning datasets forming the sub-learning dataset S will be referred to as S ₁ , . . . , S _K . Here, K is the number of sub-learning data sets in groups that constitute the sub-learning data set S (that is, the second number of groups). A specific example of the sub-learning data set S is shown in FIG. The sub-learning data set S shown in FIG. 9 is composed of group-based sub-learning data sets S ₁ , . . . , _SK . Furthermore, the group-based sub-learning data set S ₁ is composed of sub-learning data (x ₁ , x ₂ , x ₃ ) to which sample numbers m ₁ to m ₁ +|S ₁ |-1 are respectively assigned. There is. Similarly, the group-based sub-learning data set S ₂ is composed of sub-learning data (x ₂ , x ₃ ) to which sample numbers m ₂ to m ₂ +|S ₂ |-1 are respectively assigned. The same applies to other group-based sub-learning data sets. Note that |S _i | is the number of sub-learning data included in the group-based sub-learning data set S _i . The case where K=5 will be described below as an example.

Next, the learning setting information creation unit 114 of the model learning management processing unit 110 creates a learning setting information table (step S103). For example, the learning setting information creation unit 114 causes the input/output unit 113 to display a screen that visualizes the learning data set D and the sub-learning data set S on a display device 402 such as a display, and also performs a learning setting information table creation operation. is received from the input device 401, and a learning setting information table is created based on this learning setting information table creation operation. However, the present invention is not limited to this, and the learning setting information creation unit 114 may, for example, create the learning setting information table using the sub-learning data set S or both the learning data set D and the sub-learning data set S. good. A specific example of the learning setting information table is shown in FIG. The learning setting information table shown in FIG. 10 is composed of learning setting information, and each learning setting information includes an abnormality diagnosis target number, an abnormality diagnosis target name, a model number, a reference variable, a condition, and a group unit. The sub-learning dataset number is included.

The abnormality diagnosis target number is a number (identification information) that identifies the abnormality diagnosis target 310. The abnormality diagnosis target name is the name of the abnormality diagnosis target 310. The model number is a number (identification information) that identifies an abnormality diagnosis model used for abnormality diagnosis of the abnormality diagnosis target 310. A reference variable is a state variable used in a condition. The conditions refer to conditions under which the abnormality diagnosis model is used. The group-based sub-learning data set number is the number (identification information) of the group-based sub-learning data set used for learning the abnormality diagnosis model.

For example, the learning setting information in the first row of the learning setting information table shown in FIG. 10 includes the abnormality diagnosis target number "1", the abnormality diagnosis target name "device A", the model number "1", The learning data set number "1" is included. This means that the name of the abnormality diagnosis target 310 with the abnormality diagnosis target number "1" is "Device A", and that the group unit sub-learning data set S1 with the group-based sub-learning data set number "_1" is used. This indicates that the abnormality diagnosis model with the model number “1” of the abnormality diagnosis target 310 is learned, and that the abnormality diagnosis model with the model number “1” is used in the abnormality diagnosis of the abnormality diagnosis target 310.

For example, the learning setting information on the second line of the learning setting information table shown in FIG. 10 includes the abnormality diagnosis target number "2", the abnormality diagnosis target name "device B", the model number "1", and It includes variables “x _r , x _r′ ”, condition “R ₁ ”, and group unit sub-learning data set number “2”. This means that the name of the abnormality diagnosis target 310 with the abnormality diagnosis target number "2" is "Device B", and that the group unit sub-learning data set S2 with the group-based sub-learning data set number "_2" is used. Learning the abnormality diagnosis model of the model number "1" of the abnormality diagnosis target 310, when the reference variables (x _r , x _r' ) satisfy the condition R ₁ , the model This indicates that the abnormality diagnosis model numbered "1" is used.

Similarly, for example, the learning setting information in the third row of the learning setting information table shown in FIG. 10 includes the abnormality diagnosis target number "2", the abnormality diagnosis target name "device B", and the model number "2". It includes reference variables “x _r , x _r′ ”, condition “R ₂ ”, and group unit sub-learning data set number “3”. This means that the name of the abnormality diagnosis target 310 with the abnormality diagnosis target number "2" is "Device B", and that the group unit sub-learning data set S3 with the group-based sub-learning data set number " ₃ " is used. Learning an abnormality diagnosis model with model number "2" of the abnormality diagnosis target 310, when the reference variables (x _r , x _r' ) satisfy the condition R ₂ , the model This indicates that the abnormality diagnosis model numbered "2" is used.

Similarly, for example, the learning setting information in the fourth line of the learning setting information table shown in FIG. 10 includes the abnormality diagnosis target number "2", the abnormality diagnosis target name "device B", and the model number "3". The reference variables "x _r , x _r' ", the condition "R ₁ and R ₂ are not satisfied", and the group-based sub-learning data set number "4" are included. This means that the name of the abnormality diagnosis target 310 with the abnormality diagnosis target number "2" is "Device B", and that the group unit sub-learning data set S4 with the group-based sub-learning data set number " ₄ " is used. To learn the abnormality diagnosis model of the model number "3" of the abnormality diagnosis target 310, and when the reference variables (x _r , x _r' ) do not satisfy either condition R ₁ or condition R ₂ , the abnormality diagnosis target 310 This indicates that the abnormality diagnosis model with model number "3" is used in the abnormality diagnosis of 310.

Similarly, for example, the learning setting information on the fifth line of the learning setting information table shown in FIG. 10 includes the abnormality diagnosis target number "3", the abnormality diagnosis target name "device C", and the model number "1". The group unit sub-learning data set number "5" is included. This means that the name of the abnormality diagnosis target 310 with the abnormality diagnosis target number "3" is "Device C", and the group unit sub-learning data set S5 with the group-based sub-learning data set number " ₅ " is used. This indicates that the abnormality diagnosis model with the model number “1” of the abnormality diagnosis target 310 is learned, and that the abnormality diagnosis model with the model number “1” is used in the abnormality diagnosis of the abnormality diagnosis target 310.

In addition, in the learning setting information table shown in FIG. 10, "Device A", "Device B", and "Device C" exist as abnormality diagnosis targets 310, and only one operation mode exists for Device A and Device C. This is an example where B has three operation modes.

As described above, the learning setting information table includes the names of one or more abnormality diagnosis models used for abnormality diagnosis of the abnormality diagnosis target 310, information indicating the group-based sub-learning dataset used for learning the abnormality diagnosis model, This data is composed of learning setting information that includes conditions for performing an abnormality diagnosis using the abnormality diagnosis model. Note that the conditions under which abnormality diagnosis is performed using the abnormality diagnosis model are not limited to specific conditions, and any conditions expressed by any formula can be set. For example, conditions such as upper and lower limits regarding the reference variable may be used, conditions such as a conditional expression regarding a predetermined function that inputs the reference variable, or conditions such as a logical expression regarding the reference variable may be used.

Next, the learning unit 115 of the model learning management processing unit 110 acquires one piece of learning setting information that has not been acquired yet from among the learning setting information stored in the learning setting information table (step S104).

Next, the learning unit 115 of the model learning management processing unit 110 uses the group-based sub-learning data set indicated by the group-based sub-learning data set number included in the learning setting information acquired in step S104 above. An abnormality diagnosis model corresponding to the abnormality diagnosis target number and model number included in the learning setting information is learned (step S105). That is, the learning unit 115 uses the group-based sub-learning data set to learn the abnormality diagnosis model of the model number used for abnormality diagnosis of the abnormality diagnosis target 310 of the abnormality diagnosis target number. The abnormality diagnosis model learned in this step is stored as a file in a storage area such as the auxiliary storage device 407, for example.

Note that the learning unit 115 may learn the abnormality diagnosis model using any predetermined method. For example, the learning unit 115 may learn the abnormality diagnosis model using a multivariate statistical process management method, or may learn the abnormality diagnosis model using a machine learning method, as in Patent Document 1 mentioned above. .

Next, the learning unit 115 of the model learning management processing unit 110 determines whether all learning setting information has been acquired from the learning setting information table (step S106). If it is not determined that all the learning setting information has been acquired from the learning setting information table, the learning unit 115 of the model learning management processing unit 110 returns to step S104 described above. As a result, steps S104 to S105 described above are repeatedly executed until all learning setting information is acquired from the learning setting information table.

On the other hand, if it is determined that all learning setting information has been acquired from the learning setting information table, the learning unit 115 of the model learning management processing unit 110 uses the learning setting information table created in step S103 above and the learning setting information table created in step S103 above. Model management information is created from each abnormality diagnosis model learned in S105 (step S107). That is, the learning unit 115 stores a model number information table storing model number information representing the number of abnormality diagnosis models of the abnormality diagnosis target 310, and model detailed information storing model detailed information representing detailed information of the abnormality diagnosis models. Create model management information consisting of tables.

A specific example of the model number information table is shown in FIG. The model number information table shown in FIG. 11 is a model number information table created from the learning setting information table shown in FIG. The model number information table shown in FIG. 11 is composed of model number information, and each model number information includes an abnormality diagnosis target number, an abnormality diagnosis target name, and the number of models.

The abnormality diagnosis target number is a number (identification information) that identifies the abnormality diagnosis target 310. The abnormality diagnosis target name is the name of the abnormality diagnosis target 310. The number of models is the number of abnormality diagnosis models that can be used for abnormality diagnosis of the abnormality diagnosis target 310.

For example, the model number information in the first line of the model number information table shown in FIG. 11 includes the abnormality diagnosis target number "1", the abnormality diagnosis target name "equipment A", and the number of models "1". There is. This means that the name of the abnormality diagnosis target 310 with the abnormality diagnosis target number "1" is "Device A", and the number of abnormality diagnosis models that can be used for abnormality diagnosis of the abnormality diagnosis target 310 is "1". represents something.

For example, the model number information in the second line of the model number information table shown in FIG. 11 includes the abnormality diagnosis target number "2", the abnormality diagnosis target name "equipment B", and the number of models "3". It is. This means that the name of the abnormality diagnosis target 310 with the abnormality diagnosis target number "2" is "Device B", and the number of abnormality diagnosis models that can be used for abnormality diagnosis of the abnormality diagnosis target 310 is "3". represents something.

Similarly, for example, the model number information in the third row of the model number information table shown in FIG. 11 includes the abnormality diagnosis target number "3", the abnormality diagnosis target name "device C", and the number of models "1". include. This means that the name of the abnormality diagnosis target 310 with the abnormality diagnosis target number "3" is "Device C", and the number of abnormality diagnosis models that can be used for abnormality diagnosis of the abnormality diagnosis target 310 is "1". represents something.

Further, a specific example of the model detailed information table is shown in FIG. The model detailed information table shown in FIG. 12 is a model detailed information table created from the learning setting information table shown in FIG. 10 and each abnormality diagnosis model learned in step S105 above. The model detailed information table shown in FIG. 12 is composed of model detailed information, and each model detailed information includes an abnormality diagnosis target number, an abnormality diagnosis name, a model number, a model file name, a reference variable, and a condition. and explanatory variables.

The abnormality diagnosis target number is a number (identification information) that identifies the abnormality diagnosis target 310. The abnormality diagnosis name is the name of the abnormality diagnosis target 310. The model number is a number (identification information) that identifies an abnormality diagnosis model used for abnormality diagnosis of the abnormality diagnosis target 310. The model file name is the name of the file in which the abnormality diagnosis model is stored. A reference variable is a state variable used in a condition. The conditions refer to conditions under which the abnormality diagnosis model is used. An explanatory variable is a state variable that the abnormality diagnosis model takes as an input.

For example, the model detailed information in the first line of the model detailed information table shown in FIG. 12 includes the abnormality diagnosis target number "1", the abnormality diagnosis target name "device A", the model number "1", and the model file name. “001_01_20220901” and explanatory variables “x ₁ , x ₂ , x ₃ ” are included. This means that the abnormality diagnosis model with the model number “1” of the abnormality diagnosis target 310 (abnormality diagnosis target name “Device A”) with the abnormality diagnosis target number “1” is stored in the file with the file name “001_01_20220901”, and the abnormality This indicates that during diagnosis, the values of state variables (x ₁ , x ₂ , x ₃ ) of diagnostic data are input and an abnormality diagnosis result is output.

For example, the model detailed information on the second line of the model detailed information table shown in FIG. 12 includes the abnormality diagnosis target number "2", the abnormality diagnosis target name "device B", the model number "1", The file name "002_01_20220901", reference variables "x _r , x _r' ", condition "R ₁ ", and explanatory variables "x ₂ , x ₃ " are included. This means that the abnormality diagnosis model with the model number “1” of the abnormality diagnosis target 310 (abnormality diagnosis target name “Device B”) with the abnormality diagnosis target number “2” is stored in the file with the file name “002_01_20220901”, and the The abnormality diagnosis model is used when the value of the _{state variable} (x _{r ,} This indicates that the value of x ₃ ) is input and the abnormality diagnosis result is output.

Similarly, for example, the model detailed information in the third row of the model detailed information table shown in FIG. 12 includes the abnormality diagnosis target number "2", the abnormality diagnosis target name "device B", and the model number "2". The model file name "002_02_20220901", reference variables "x _r , x _r' ", condition "R ₂ ", and explanatory variables "x ₂ , x ₃ " are included. This means that the abnormality diagnosis model with the model number “2” of the abnormality diagnosis target 310 (abnormality diagnosis target name “Device B”) with the abnormality diagnosis target number “2” is stored in the file with the file name “002_02_20220901”, and the The abnormality diagnosis model is used when the value of the _{state variable} (x _{r ,} This indicates that the value of x ₃ ) is input and the abnormality diagnosis result is output.

Similarly, for example, the model detailed information on the fourth line of the model detailed information table shown in FIG. 12 includes the abnormality diagnosis target number "2", the abnormality diagnosis target name "device B", and the model number "3". The model file name "002_03_20220901", reference variables "x _r , x _r' ", conditions "R ₁ and R ₂ are not satisfied", and explanatory variables "x ₂ , x ₃ " are included. This means that the abnormality diagnosis model with the model number "3" of the abnormality diagnosis target 310 (abnormality diagnosis target name "Device B") with the abnormality diagnosis target number "2" is stored in the file with the file name "002_03_20220901", and the The abnormality diagnosis model is used when the value of the state variable (x _r , x _r' ) of the diagnostic data does not satisfy either condition R ₁ or condition R ₂ . This indicates that the values of the state variables (x ₂ , x ₃ ) are input and an abnormality diagnosis result is output.

Similarly, for example, the model detailed information on the fifth line of the model detailed information table shown in FIG. 12 includes the abnormality diagnosis target number "3", the abnormality diagnosis target name "device C", and the model number "1". The model file name “003_01_20220901” and explanatory variables “x ₂ , x ₃ , x ₄ ” are included. This means that the abnormality diagnosis model with the model number "1" of the abnormality diagnosis target 310 (abnormality diagnosis target name "Device C") with the abnormality diagnosis target number "3" is stored in the file with the file name "003_01_20220901", and the abnormality This indicates that during diagnosis, the values of state variables (x ₂ , x ₃ , x ₄ ) of diagnostic data are input and an abnormality diagnosis result is output.

As described above, the model management information includes a model number information table storing model number information representing the number of abnormality diagnosis models of the abnormality diagnosis target 310, and model detailed information representing detailed information of the abnormality diagnosis models. It consists of a model detailed information table. This makes it possible to manage various information such as the number of abnormality diagnosis models used for abnormality diagnosis of each abnormality diagnosis target 310, conditions under which the abnormality diagnosis models are used, and the like.

Then, the learning unit 115 of the model learning management processing unit 110 uses the model management information created in step S107 above and each abnormality diagnosis model (more precisely, each abnormality diagnosis model learned in step S105 above). and the stored files) to the abnormality diagnosis processing unit 210 of the abnormality diagnosis device 20 (step S108). Thereby, the model management information and each abnormality diagnosis model are stored in the storage area of the auxiliary storage device 507 of the abnormality diagnosis apparatus 20 by the abnormality diagnosis processing unit 210.

<Abnormality diagnosis processing>
The abnormality diagnosis process according to this embodiment will be described below with reference to FIG. 13. This abnormality diagnosis process is repeatedly executed, for example, every time new driving data is stored in the driving database 220.

First, the diagnostic data acquisition unit 211 of the abnormality diagnosis processing unit 210 acquires new driving data stored in the driving database 220 as diagnostic data (step S201).

Next, the abnormality diagnosis unit 212 of the abnormality diagnosis processing unit 210 refers to the model management information and identifies an abnormality diagnosis model to be used for abnormality diagnosis of each abnormality diagnosis target 310 (step S202). The abnormality diagnosis unit 212 may specify an abnormality diagnosis model to be used for abnormality diagnosis of each abnormality diagnosis target 310, for example, according to steps 21 and 22 below.

Step 21: The abnormality diagnosis unit 212 refers to the model number information table and identifies the number of models for each abnormality diagnosis target 310.

For example, the abnormality diagnosis unit 212 refers to the model number information table shown in FIG. , the model number of the abnormality diagnosis target 310 with the abnormality diagnosis target number "2" (abnormality diagnosis target name "Device B") is "3", and the number of models of the abnormality diagnosis target 310 with the abnormality diagnosis target number "3" (the abnormality diagnosis target name "Device B") is The number of models of "C") is specified as "1".

Step 22: The abnormality diagnosis unit 212 refers to the model detailed information table and determines, for each abnormality diagnosis target 310, one abnormality diagnosis model (or the number of models of the abnormality diagnosis target 310) used for the abnormality diagnosis of the abnormality diagnosis target 310. It may be one or more of the following abnormality diagnosis models.

For example, the abnormality diagnosis unit 212 refers to the model detailed information table shown in FIG. The abnormality diagnosis model with model number "1" (model file name "001_01_20220901") is specified as the model.

Further, for example, the abnormality diagnosis unit 212 refers to the model detailed information table shown in FIG. As the abnormality diagnosis model, one of the abnormality diagnosis models with model numbers "1" to "3" is specified depending on which condition is satisfied. That is, for example _, when the values of the state variables (x _{r ,} 002_01_20220901"), and when the value of the state variable (x _r , x _r' ) of the diagnostic data satisfies condition R ₂ , the abnormality diagnosis model with model number "2" (model file name "002_02_20220901") is specified. _is specified _, and if the values of the state variables (x _{r ,} 002_03_20220901").

Furthermore, for example, the abnormality diagnosis unit 212 refers to the model detailed information table shown in FIG. The abnormality diagnosis model with model number "1" (model file name "003_01_20220901") is specified as the abnormality diagnosis model.

Note that it is possible for a certain abnormality diagnosis target 310 to use a plurality of abnormality diagnosis models, and the conditions under which at least some of the plurality of abnormality diagnosis models are used are mutually exclusive. If not, one or more abnormality diagnosis models less than or equal to the number of models of the abnormality diagnosis target 310 may be specified.

Next, the abnormality diagnosis unit 212 of the abnormality diagnosis processing unit 210 refers to the model management information, and for each abnormality diagnosis target 310, the abnormality diagnosis model specified to be used for the abnormality diagnosis of the abnormality diagnosis target 310, and the An abnormality diagnosis is performed using the data for the processing (step S203). That is, for each abnormality diagnosis target 310, the abnormality diagnosis unit 212 selects the abnormality diagnosis model from among the values of the state variables of the diagnostic data for the abnormality diagnosis model specified to be used for the abnormality diagnosis of the abnormality diagnosis target 310. Enter the values of the explanatory variables that will be taken as input. As a result, for each abnormality diagnosis target 310, an abnormality diagnosis result is output from the abnormality diagnosis model used for the abnormality diagnosis of the abnormality diagnosis target 310.

Then, the input/output unit 213 outputs the abnormality diagnosis results output from each abnormality diagnosis model in step S203 above (that is, the abnormality diagnosis results of each abnormality diagnosis target 310) to a predetermined output destination (step S204). . For example, the input/output unit 213 displays a screen that visualizes the abnormality diagnosis results of each abnormality diagnosis target 310 on the display device 502 such as a display. However, this is just an example, and the input/output unit 213 may output and save the abnormality diagnosis results of each abnormality diagnosis target 310 to a storage area such as the auxiliary storage device 507, or It may also be output to a terminal device used by an operation operator, administrator, or the like. Further, the input/output unit 213 may output the abnormality diagnosis results of each abnormality diagnosis target 310 to the control device 50. By outputting the abnormality diagnosis results of each abnormality diagnosis target 310 to the control device 50, for example, when an abnormality occurs in a certain abnormality diagnosis target 310, the production equipment 30 or the abnormality diagnosis target 310 constituting it can be Controls such as stop and degenerate operation can be realized.

<Summary>
As described above, the abnormality diagnosis system 1 according to the present embodiment creates model management information that is management information of each abnormality diagnosis model when learning the abnormality diagnosis model of the abnormality diagnosis target 310 that constitutes the production equipment 30. do. Therefore, according to the abnormality diagnosis system 1 according to the present embodiment, it is not necessary to manage each abnormality diagnosis model individually, and it is possible to manage it using model management information, so that management costs can be reduced. becomes. Note that, for example, when a new anomaly diagnosis model is added or an existing anomaly diagnosis model is deleted or updated, the model management information only needs to be updated. The associated costs can also be reduced.

Further, the abnormality diagnosis system 1 according to the present embodiment can identify an abnormality diagnosis model to be used for abnormality diagnosis of each abnormality diagnosis target 310 by using model management information when diagnosing each abnormality diagnosis target 310. It becomes possible. Therefore, according to the abnormality diagnosis system 1 according to the present embodiment, it is necessary to individually set the conditions etc. when the plurality of abnormality diagnosis models are used for the abnormality diagnosis target 310 for which a plurality of abnormality diagnosis models can be used. This eliminates this problem, making it possible to reduce the setup cost.

<Modified example>
In the above embodiment, the abnormality diagnosis results of each abnormality diagnosis model are output to a predetermined output destination, but for example, as in Patent Document 1 mentioned above, at least one of the abnormality diagnosis results of each abnormality diagnosis model is The abnormality diagnosis result may be output by integrating the abnormality diagnosis results of the parts (in particular, the abnormality diagnosis results of the plurality of abnormality diagnosis models respectively corresponding to the plurality of operation modes of the same abnormality diagnosis target 310). Note that as a method for integrating the abnormality diagnosis results, a method similar to the method described in Patent Document 1 mentioned above may be used.

The present invention is not limited to the above-described specifically disclosed embodiments, and various modifications and changes, combinations with known techniques, etc. can be made without departing from the scope of the claims. be.

[References]
Reference 1: Statistical process management using process chemometrics, System/Control/Information, Vol.48, No.5, pp.165-170, 2004.

1 Abnormality Diagnosis System 10 Model Learning Management Device 20 Abnormality Diagnosis Device 30 Production Equipment 40 Measuring Instrument 50 Control Device 110 Model Learning Management Processing Unit 111 Learning Data Acquisition Unit 112 Learning Data Classification Unit 113 Input/Output Unit 114 Learning Setting Information Creation Unit 115 Learning section 210 Abnormality diagnosis processing section 211 Diagnostic data acquisition section 212 Abnormality diagnosis section 213 Input/output section 220 Operation database 310 Abnormality diagnosis target 401 Input device 402 Display device 403 External I/F
403a Recording medium 404 Communication I/F
405 RAM
406 ROM
407 Auxiliary storage device 408 Processor 409 Bus 501 Input device 502 Display device 503 External I/F
503a Recording medium 504 Communication I/F
505 RAM
506 ROM
507 Auxiliary storage device 508 Processor 509 Bus

Claims (7)

An abnormality diagnosis system that performs abnormality diagnosis of equipment composed of multiple abnormality diagnosis targets in which one or more operation modes exist,
a first acquisition unit configured to acquire a first data set composed of learning data representing past operation data of the equipment;
a classification unit configured to create a second data set in which the learning data included in the first data set is classified into groups for each operation mode of each of the plurality of abnormality diagnosis targets;
Learning an abnormality diagnosis model for each operation mode of each of the plurality of abnormality diagnosis targets based on either the second data set or both the first data set and the second data set. a first creation unit configured to create learning setting information for
For each group, among the learning data included in the second data set, state variable values of at least some of the learning data classified into the group are used to determine the abnormality diagnosis target and the data corresponding to the group. a learning unit configured to learn an abnormality diagnosis model of the operating mode;
A second device configured to create model management information for managing an abnormality diagnosis model for each operation mode of each of the plurality of abnormality diagnosis targets based on the learning setting information and the abnormality diagnosis model. The creation department of
has
The learning setting information includes identification information for identifying each of the plurality of abnormality diagnosis targets, and learning used for learning an abnormality diagnosis model in the operation mode for each operation mode of the abnormality diagnosis target identified by the identification information. An abnormality diagnosis system that includes at least information representing a group in which the data for use is included, and conditions under which the abnormality diagnosis model is used. The second creation unit is
2. The computer according to claim 1, wherein model management information is created for each operation mode of each of the plurality of abnormality diagnosis targets, the model management information including at least a condition when the abnormality diagnosis model of the operation mode is used. Abnormality diagnosis system. The second creation unit is
The number of abnormality diagnosis models that can be used by each of the plurality of abnormality diagnosis targets, the conditions under which the abnormality diagnosis model is used, the state variables for determining the conditions, and the input of the abnormality diagnosis model. The abnormality diagnosis system according to claim 2, wherein the abnormality diagnosis system is configured to create model management information that includes state variables taken as . a second acquisition unit configured to acquire diagnostic data representing the latest operational data of the equipment;
It is configured to diagnose whether or not an abnormality has occurred in each of the plurality of abnormality diagnosis targets using the diagnostic data and an abnormality diagnosis model for each operation mode of each of the plurality of abnormality diagnosis targets. It has an abnormality diagnosis department,
The abnormality diagnosis section includes:
For each of the plurality of abnormality diagnosis targets, use an abnormality diagnosis model that satisfies the conditions among the abnormality diagnosis models for each operation mode of the abnormality diagnosis target to determine whether or not the abnormality has occurred in the abnormality diagnosis target. The abnormality diagnosis system according to claim 2 or 3, configured to perform diagnosis. An abnormality diagnosis device that performs abnormality diagnosis of equipment composed of multiple abnormality diagnosis targets in which one or more operation modes exist,
a first acquisition unit configured to acquire a first data set composed of learning data representing past operation data of the equipment;
a classification unit configured to create a second data set in which the learning data included in the first data set is classified into groups for each operation mode of each of the plurality of abnormality diagnosis targets;
Learning an abnormality diagnosis model for each operation mode of each of the plurality of abnormality diagnosis targets based on either the second data set or both the first data set and the second data set. a first creation unit configured to create learning setting information for
For each group, among the learning data included in the second data set, state variable values of at least some of the learning data classified into the group are used to determine the abnormality diagnosis target and the data corresponding to the group. a learning unit configured to learn an abnormality diagnosis model of the operating mode;
A second device configured to create model management information for managing an abnormality diagnosis model for each operation mode of each of the plurality of abnormality diagnosis targets based on the learning setting information and the abnormality diagnosis model. The creation department of
has
The learning setting information includes identification information for identifying each of the plurality of abnormality diagnosis targets, and learning used for learning an abnormality diagnosis model in the operation mode for each operation mode of the abnormality diagnosis target identified by the identification information. An abnormality diagnosing device that includes at least information representing a group in which the data for the abnormality diagnosis is included, and conditions under which the abnormality diagnosis model is used. An abnormality diagnosis method executed by an abnormality diagnosis system for diagnosing an abnormality of equipment configured with a plurality of abnormality diagnosis targets in which one or more operation modes exist, the method comprising:
a first acquisition procedure of acquiring a first data set composed of learning data representing past operation data of the equipment;
a classification procedure for creating a second data set in which learning data included in the first data set is classified into groups for each operation mode of each of the plurality of abnormality diagnosis targets;
Learning an abnormality diagnosis model for each operation mode of each of the plurality of abnormality diagnosis targets based on either the second data set or both the first data set and the second data set. a first creation step of creating learning setting information for
For each group, among the learning data included in the second data set, state variable values of at least some of the learning data classified into the group are used to determine the abnormality diagnosis target and the data corresponding to the group. A learning procedure for learning an abnormality diagnosis model of an operation mode;
a second creation step of creating model management information for managing an abnormality diagnosis model for each operation mode of each of the plurality of abnormality diagnosis targets based on the learning setting information and the abnormality diagnosis model;
contains ,
The learning setting information includes identification information for identifying each of the plurality of abnormality diagnosis targets, and learning used for learning an abnormality diagnosis model in the operation mode for each operation mode of the abnormality diagnosis target identified by the identification information. An abnormality diagnosis method , the method comprising at least information representing a group in which data for the abnormality diagnosis is included, and conditions under which the abnormality diagnosis model is used. An abnormality diagnosis system that diagnoses abnormalities in equipment that is composed of multiple abnormality diagnosis targets that have one or more operation modes,
a first acquisition procedure of acquiring a first data set composed of learning data representing past operation data of the equipment;
a classification procedure for creating a second data set in which learning data included in the first data set is classified into groups for each operation mode of each of the plurality of abnormality diagnosis targets;
Learning an abnormality diagnosis model for each operation mode of each of the plurality of abnormality diagnosis targets based on either the second data set or both the first data set and the second data set. a first creation step of creating learning setting information for
For each group, among the learning data included in the second data set, state variable values of at least some of the learning data classified into the group are used to determine the abnormality diagnosis target and the data corresponding to the group. A learning procedure for learning an abnormality diagnosis model of an operation mode;
a second creation step of creating model management information for managing an abnormality diagnosis model for each operation mode of each of the plurality of abnormality diagnosis targets based on the learning setting information and the abnormality diagnosis model;
run the
The learning setting information includes identification information for identifying each of the plurality of abnormality diagnosis targets, and learning used for learning an abnormality diagnosis model in the operation mode for each operation mode of the abnormality diagnosis target identified by the identification information. A program that includes at least information representing a group in which data for use is included, and conditions under which the abnormality diagnosis model is used.

Images