JP2020107027A - Abnormality diagnosis device, abnormality diagnosis method, abnormality diagnosis program, and recording medium - Google Patents

Abstract

To provide an abnormality diagnosis device, an abnormality diagnosis method, an abnormality diagnosis program, and a recording medium capable of improving diagnosis accuracy.SOLUTION: An abnormality diagnosis device 1 comprises: an acquisition unit 11 that acquires diagnosis target data from a mechanical system 2; an extraction unit 12 that determines an extraction condition using the diagnosis target data and condition setting information, and extracts learning data from extraction source data including a plurality of data sets; a forming unit 13 that forms learning information from the learning data; and a diagnosis unit 14 that determines whether or not the diagnosis target data is abnormal based on the learning information and a diagnostic threshold value according to a total number of the extracted data sets which are extracted as the learning data. Each of the plurality of data sets includes parameter values of a plurality of items when the mechanical system 2 is in a normal state and the extraction unit 12 extracts, as the learning data, a group of data sets satisfying the extraction condition from among the plurality of data sets included in the extraction source data.SELECTED DRAWING: Figure 1

Description

The present disclosure relates to an abnormality diagnosis device, an abnormality diagnosis method, an abnormality diagnosis program, and a recording medium.

In a mechanical system such as a plant in which multiple devices operate, the status of the system is measured using multiple sensors, and the measured data can be used to perform abnormality diagnosis such as whether a problem has occurred in the mechanical system. It is common. In recent years, analysis methods using pattern recognition technology have been studied for the purpose of highly accurately diagnosing abnormalities in mechanical systems.

For example, in Patent Document 1, a data set satisfying the extraction condition is extracted as learning data from a plurality of data sets indicating past mechanical system states for the purpose of performing abnormality diagnosis according to a change in the mechanical system status. , An abnormality diagnosis device that performs abnormality diagnosis based on learning information created from learning data. In this abnormality diagnosis device, by combining the parameter value included in the diagnosis target data and the extraction condition information for determining the extraction condition of the learning data, the extraction condition of the learning data is determined, and thus the diagnosis target data is obtained. Learning data is dynamically selected accordingly.

JP, 2017-102826, A

There is a method such as MT method (Mahalanobis-Taguchi Method) which performs abnormality diagnosis by calculating an abnormality degree score of data to be diagnosed and comparing a predetermined diagnostic threshold value with the abnormality degree score. .. In the abnormality diagnosing device described in Patent Document 1, since the data set that satisfies the extraction condition is used as the learning data, the number of data sets included in the learning data may change according to the extraction condition. In general, when the number of data sets included in the learning data is small, the variation in the abnormality score when the normal diagnosis target data is diagnosed is large. Therefore, when the abnormality diagnosis apparatus described in Patent Document 1 performs abnormality diagnosis using the abnormality degree score, there is a possibility that erroneous detection may occur in which normal diagnosis target data is diagnosed as an abnormality. If the diagnosis threshold is set to a large value in order to reduce erroneous detection, there is a possibility that undetected abnormal data to be diagnosed may be detected as normal.

The present disclosure describes an abnormality diagnosis device, an abnormality diagnosis method, an abnormality diagnosis program, and a recording medium that can improve diagnosis accuracy.

An abnormality diagnosis device according to one aspect of the present disclosure is a device that performs an abnormality diagnosis on a diagnosis target system using a pattern recognition method. This abnormality diagnosis device includes an acquisition unit that acquires diagnosis target data including parameter values of a plurality of items from a diagnosis target system, a first storage unit that stores extraction source data including a plurality of data sets, and a diagnosis target data. An extraction unit that determines the extraction condition using the condition setting information for determining the extraction condition and extracts learning data from the extraction source data, and a creation unit that creates learning information used for the pattern recognition method from the learning data. A diagnosis unit that determines whether or not the diagnosis target data is abnormal based on the learning information and a diagnosis threshold value according to the number of extracted data sets extracted as learning data by the extraction unit. Each of the plurality of data sets includes parameter values of a plurality of items when the system to be diagnosed is in a normal state. The extraction unit extracts, as learning data, a group of data sets satisfying the extraction condition from the plurality of data sets included in the extraction source data.

According to the present disclosure, diagnostic accuracy can be improved.

FIG. 1 is a diagram showing functional blocks of an abnormality diagnosis device according to an embodiment. FIG. 2 is a diagram showing an example of the threshold table. FIG. 3 is a diagram showing the distribution of the abnormality score for each number of extractions. FIG. 4 is a flowchart showing a series of processes of the abnormality diagnosis method performed by the abnormality diagnosis device of FIG. FIG. 5 is a flowchart showing the update processing of the extraction source data performed by the abnormality diagnosis device of FIG. FIG. 6 is a flowchart showing a threshold table setting process performed by the abnormality diagnosis device of FIG. FIG. 7 is a diagram showing the structure of the abnormality diagnosis program. FIG. 8A is a diagram for explaining the diagnosis threshold value in the abnormality diagnosis device of FIG. 1. FIG. 8B is a diagram for explaining a diagnosis threshold in the abnormality diagnosis device according to the comparative example. FIG. 9 is a diagram for explaining a modified example of the threshold table setting process.

Since the embodiments according to the present disclosure described below are examples for explaining the present invention, the present invention should not be limited to the following contents.

[1] Outline of Embodiment An abnormality diagnosis apparatus according to one aspect of the present disclosure is an apparatus that performs an abnormality diagnosis on a diagnosis target system using a pattern recognition method. This abnormality diagnosis device includes an acquisition unit that acquires diagnosis target data including parameter values of a plurality of items from a diagnosis target system, a first storage unit that stores extraction source data including a plurality of data sets, and a diagnosis target data. An extraction unit that determines the extraction condition using the condition setting information for determining the extraction condition and extracts learning data from the extraction source data, and a creation unit that creates learning information used for the pattern recognition method from the learning data. A diagnosis unit that determines whether or not the diagnosis target data is abnormal based on the learning information and a diagnosis threshold value according to the number of extracted data sets extracted as learning data by the extraction unit. Each of the plurality of data sets includes parameter values of a plurality of items when the system to be diagnosed is in a normal state. The extraction unit extracts, as learning data, a group of data sets satisfying the extraction condition from the plurality of data sets included in the extraction source data.

An abnormality diagnosis method according to another aspect of the present disclosure is a method executed by an abnormality diagnosis device that performs an abnormality diagnosis on a system to be diagnosed using a pattern recognition method. This abnormality diagnosis method determines extraction conditions using a step of acquiring diagnosis target data including parameter values of a plurality of items from a diagnosis target system and condition setting information for determining the diagnosis target data and extraction conditions. A step of extracting learning data from extraction source data including a plurality of data sets, a step of creating learning information used for the pattern recognition method from the learning data, learning information, and a data set extracted as the learning data And a step of determining whether or not the diagnosis target data is abnormal based on the diagnosis threshold value according to the number of extracted cases. Each of the plurality of data sets includes parameter values of a plurality of items when the system to be diagnosed is in a normal state. In the extracting step, a group of data sets satisfying the extraction condition is extracted as learning data from the plurality of data sets included in the extraction source data.

An abnormality diagnosis program according to yet another aspect of the present disclosure is a program that causes a computer to execute an abnormality diagnosis related to a diagnosis target system using a pattern recognition method. This abnormality diagnosis program determines an extraction condition using a step of acquiring diagnosis target data including parameter values of a plurality of items from a diagnosis target system and condition setting information for determining the diagnosis target data and the extraction condition. A step of extracting learning data from extraction source data including a plurality of data sets, a step of creating learning information used for the pattern recognition method from the learning data, learning information, and a data set extracted as the learning data And a step of determining whether or not the diagnosis target data is abnormal based on the diagnosis threshold value according to the number of extracted cases of. Each of the plurality of data sets includes parameter values of a plurality of items when the system to be diagnosed is in a normal state. In the extracting step, a group of data sets satisfying the extraction condition is extracted as learning data from the plurality of data sets included in the extraction source data.

A recording medium according to yet another aspect of the present disclosure is a computer-readable recording medium that records an abnormality diagnosis program that causes a computer to execute an abnormality diagnosis of a diagnosis target system using a pattern recognition method. The abnormality diagnosis program recorded in this recording medium uses a step of acquiring diagnosis target data including parameter values of a plurality of items from the diagnosis target system, and condition setting information for determining the diagnosis target data and extraction conditions. Determining the extraction conditions, extracting learning data from the extraction source data including multiple data sets, creating learning information used for the pattern recognition method from the learning data, learning information, and learning data. And a step of determining whether or not the diagnosis target data is abnormal based on the diagnostic threshold value according to the number of extracted data sets extracted as. Each of the plurality of data sets includes parameter values of a plurality of items when the system to be diagnosed is in a normal state. In the extracting step, a group of data sets satisfying the extraction condition is extracted as learning data from the plurality of data sets included in the extraction source data.

In the abnormality diagnosis device, the abnormality diagnosis method, the abnormality diagnosis program, and the recording medium, the extraction condition is determined using the diagnosis target data and the condition setting information for determining the extraction condition, and a plurality of extraction source data are included. Of the data sets, a group of data sets that satisfy the extraction condition is extracted as learning data, and learning information used for the pattern recognition method is created from the learning data. When the diagnosis target data is diagnosed using the learning information created in this way, the range of values that the abnormality score can take varies depending on the number of data sets extracted, even if the diagnosis target data is normal. .. On the other hand, in the abnormality diagnosis device, the abnormality diagnosis method, the abnormality diagnosis program, and the recording medium, based on the learning information and the diagnosis threshold value according to the number of extracted data sets extracted as learning data, the diagnosis target It is determined whether the data is abnormal. For example, in consideration of the possible range of the abnormality score of the normal diagnosis target data, the diagnosis threshold is set according to the number of extracted data sets included in the learning data, so that the normal diagnosis target data is diagnosed as abnormal. It is possible to reduce erroneous detection that occurs, and undetected occurrence that abnormal diagnosis target data is diagnosed as normal. As a result, the diagnostic accuracy can be improved.

The diagnostic threshold may be set to a larger value as the number of extracted cases is smaller. In general, the smaller the number of extracted data sets included in the learning data, the larger the variation in the abnormality score when the normal diagnosis target data is diagnosed. Therefore, the smaller the number of data sets extracted is, the larger the diagnostic threshold value is set, so that it is possible to reduce the occurrence of erroneous detection. In addition, the larger the number of data sets extracted, the smaller the diagnostic threshold is set, so that it is possible to reduce the number of undetected cases.

The abnormality diagnosis device may further include a second storage unit that stores a plurality of threshold values respectively associated with the number of data sets. The diagnostic unit may select a threshold value associated with the number of extracted items as a diagnostic threshold value from a plurality of threshold values. In this case, the diagnostic threshold value is set only by acquiring the threshold value associated with the number of extracted items from the second storage unit, so that the setting of the diagnostic threshold value can be simplified.

The abnormality diagnosis device may further include a setting unit that sets a plurality of threshold values. The setting unit sets a partial extraction source data that is a first subset of the plurality of data sets included in the extraction source data and a verification data group that is a second subset of the plurality of data sets included in the extraction source data. You may prepare. The setting unit may extract the threshold learning data, which is the learning data for threshold setting, from the partial extraction source data by using the verification data included in the verification data group and the condition setting information. Threshold learning information, which is learning information for setting a threshold, may be created. The setting unit may calculate the abnormality score of the verification data using the threshold learning information, and may set a plurality of thresholds based on the abnormality score and the number of data sets included in the threshold learning data. Good. Each of the plurality of data sets included in the extraction source data includes parameter values of a plurality of items when the diagnosis target system is in a normal state. Therefore, by calculating the abnormality degree score using the extraction source data, it is possible to grasp the range that the abnormality degree score of the normal diagnosis target data can take. With this, it is possible to set a plurality of threshold values that are respectively associated with the number of data sets based on the range in which the abnormality score of the normal diagnosis target data can be obtained, and thus it is possible to optimize the diagnosis threshold value. Becomes As a result, the diagnostic accuracy can be further improved.

The setting unit may further prepare an anomaly verification data group including a plurality of anomaly data sets, and a first distribution acquired based on the verification data group and a second distribution acquired based on the anomaly verification data group, A plurality of thresholds may be set based on Each of the plurality of abnormal data sets may include parameter values of a plurality of items when the system to be diagnosed is in an abnormal state. Each of the first distribution and the second distribution may indicate a possible range of the abnormality score for each number of extractions. In this case, the threshold value is set for each number of extractions in consideration of not only the range of the abnormality degree score of the normal diagnosis target data but also the range of the abnormality degree score of the abnormal diagnosis target data. Therefore, the diagnostic threshold is further optimized, and the diagnostic accuracy can be further improved.

The abnormality diagnosis device may further include an update unit that updates the extraction source data stored in the first storage unit. The update unit may update the extraction source data by adding a normal data set to the extraction source data. The setting unit may reset a plurality of thresholds according to the update of the extraction source data. In this case, a plurality of thresholds are reset according to the addition of the normal data set. As a result, the diagnostic threshold value can be selected from a plurality of threshold values that reflect the latest status of the system to be diagnosed, so that it is possible to further improve the diagnostic accuracy.

[2] Example of Embodiment Hereinafter, an embodiment of the present disclosure will be described with reference to the drawings. In the description of the drawings, the same elements will be denoted by the same reference symbols, without redundant description.

FIG. 1 is a diagram showing functional blocks of an abnormality diagnosis device according to an embodiment. The abnormality diagnosis device 1 shown in FIG. 1 is a device for performing an abnormality diagnosis related to the mechanical system 2 using a pattern recognition method. The abnormality diagnosis device 1 is communicatively connected to the mechanical system 2 and the external device 3. The pattern recognition method is a method of determining whether or not the pattern indicated by the diagnosis target data that is the target of abnormality diagnosis is normal based on the learning information formed by the learning data. Examples of pattern recognition methods include MT method (Mahalanobis-Taguchi Method), RT method (Recognition-Taguchi Method), error pressure method (standardized error pressure method), RE method (reconstruction error method), SBL (Sparse Bayesian Learning), Examples include principal component analysis, multiple regression analysis, and logistic regression analysis (multivariate analysis). The pattern recognition method is not limited to these.

The mechanical system 2 is a system to be diagnosed. Examples of the mechanical system 2 include mechanical systems such as a gas turbine, an aviation engine, and a vacuum furnace. A plurality of sensors are attached to the mechanical system 2 in order to confirm the operating status of the mechanical system 2. The mechanical system 2 transmits a data set including parameter values of a plurality of items to the abnormality diagnosis device 1 as diagnosis target data. The dataset is also referred to as sample data.

The item is a characteristic amount of the mechanical system 2. Examples of the items include control values, command values, and response values of a plurality of types of devices (valves, pumps, and the like) included in the mechanical system 2, and each detected by a plurality of sensors provided in the mechanical system 2. The sensor value is included. The number of items of the parameter value is, for example, several hundreds or more.

The sample data (data set) may include not only parameter values that change depending on the internal situation of the mechanical system 2 but also parameter values that change depending on the external situation of the mechanical system 2. The parameter value that changes depending on the external situation of the mechanical system 2 is a value designated by an external device or the like, and a value that indicates information related to the external environment, and is different from the value derived from the control in the mechanical system 2. Examples of such parameter values include the temperature around the mechanical system 2 and the output set value of a specific device. On the other hand, the parameter value that changes according to the internal situation of the mechanical system 2 is a parameter value that changes according to the operation of the mechanical system 2. Examples of such parameter values include sensor values of sensors provided in the mechanical system 2. For the abnormality diagnosis of the mechanical system 2, at least a parameter value that changes depending on the internal situation is used.

The external device 3 is composed of, for example, a display and has a function of outputting the diagnosis result of the abnormality diagnosis device 1.

The abnormality diagnosis device 1 physically includes a processor 10 and a storage device 20.

The storage device 20 is configured by a recording medium capable of reading and writing data, such as a RAM (Random Access Memory), a semiconductor memory, and a hard disk device. The storage device 20 includes an accumulated data storage unit 21 (first storage unit), a learning information storage unit 22, a threshold storage unit 23 (second storage unit), and an abnormality diagnosis program P.

The accumulated data storage unit 21 stores extraction source data which is a group of data sets that may be used as learning data. The extraction source data includes a plurality of data sets. Each of the plurality of data sets is sample data when the mechanical system 2 is in a normal state, and includes parameter values of a plurality of items in the normal state. Each of the plurality of data sets is sample data transmitted in the past from the mechanical system 2. Each of the plurality of data sets includes time information indicating the time when the data set was acquired, a status flag indicating whether the data set is normal or abnormal, and the like.

The extraction source data may include not only the sample data transmitted from the mechanical system 2 but also other data. That is, the extraction source data is data acquired from a mechanical system different from the mechanical system 2 (for example, equipment of the same system existing at another site) and specific environmental conditions in the system imitating the mechanical system 2. It may include simulation results and the like.

The learning information storage unit 22 temporarily holds the learning information created by the creating unit 13 described below. The learning information is information used for the pattern recognition method. Although details will be described later, in the abnormality diagnosis device 1, learning information is created every time the diagnosis target data described later is acquired from the mechanical system 2. The learning information storage unit 22 temporarily holds the learning information created each time an abnormality diagnosis is made, and erases the learning information used for the abnormality diagnosis when a series of processes relating to the abnormality diagnosis is completed.

The threshold storage unit 23 stores a plurality of thresholds that are associated with the number of extracted data sets. Specifically, the threshold storage unit 23 stores a threshold table set by the setting unit 17 described later. As shown in FIG. 2, in the threshold table, the number of extracted data sets and the threshold are associated with each other. In FIG. 2, each threshold value is associated with the range of the number of extracted items. One threshold value is assigned to each number of extractions.

Examples of the processor 10 include a CPU (Central Processing Unit), a microcontroller, and a DSP (Digital Signal Processor). The processor 10 may be a single processor or a multiprocessor. Functionally, the processor 10 includes an acquisition unit 11, an extraction unit 12, a creation unit 13, a diagnosis unit 14, an output unit 15, an update unit 16, and a setting unit 17. The acquisition unit 11, the extraction unit 12, the creation unit 13, the diagnosis unit 14, the output unit 15, the update unit 16, and the setting unit by the processor 10 reading and executing the abnormality diagnosis program P stored in the storage device 20. Each function of 17 is realized. The specific configuration of the abnormality diagnosis program P will be described later.

The acquisition unit 11 acquires the diagnosis target data from the mechanical system 2. The acquisition unit 11 outputs the acquired diagnosis target data to the extraction unit 12 and the diagnosis unit 14.

The extraction unit 12 extracts learning data from the extraction source data based on the diagnosis target data received from the acquisition unit 11 and the condition setting information preset in the extraction unit 12. Specifically, the extraction unit 12 determines the extraction condition of the learning data using the diagnosis target data and the condition setting information. The extraction unit 12 reads the extraction source data stored in the accumulated data storage unit 21 from the accumulated data storage unit 21, and extracts learning data from the extraction source data based on the extraction condition. Specifically, the extraction unit 12 extracts, from the extraction source data, a group of data sets satisfying the extraction condition among the plurality of data sets included in the extraction source data as learning data. The extraction unit 12 outputs the extracted learning data to the creation unit 13.

Here, a method of determining the extraction condition will be described in detail. The condition setting information is information for determining the extraction condition of the learning data, and indicates the condition regarding one or more items included in the plurality of items forming the data set. The condition setting information includes, for example, an item ID (Identifier) capable of uniquely identifying the item, and condition information indicating a condition for the parameter value of the item indicated by the item ID. The extraction unit 12 acquires the parameter value of the item indicated by the item ID included in the condition setting information from the diagnosis target data, and extracts the extraction condition using the parameter value and the condition indicated by the condition information included in the condition setting information. To decide.

For example, assume that the item indicated by the item ID is “intake air temperature” and the condition indicated by the condition information is “±2 degrees”. When the intake air temperature (parameter value) of the diagnosis target data is “20 degrees”, the extraction condition is determined to be “the intake air temperature (parameter value) is 18 degrees or more and 22 degrees or less”. To be done. For example, it is assumed that the item indicated by the item ID is “setting output” and the condition indicated by the condition information is “±1 MW (megawatts)”. When the output (parameter value) of the diagnosis target data is “10 MW”, the extraction condition is determined to be “the set output (parameter value) is 9 MW or more and 11 MW or less”.

As described above, the condition setting information is information that specifically shows a part related to the condition for determining the extraction condition, and by using the specific parameter value included in the diagnosis target data, This is information for determining extraction conditions. It is also possible to set the extraction condition such as “the parameter value of the parameter C is 10 to 30” without using the specific parameter value included in the diagnosis target data. However, since the learning data is extracted without considering the situation of the diagnosis target data acquired from the mechanical system 2, the learning information having low relevance to the diagnosis target data is created. Therefore, the abnormality diagnosis apparatus 1 adopts a configuration in which the parameter value included in the diagnosis target data and the condition setting information for determining the extraction condition of the learning data are combined to determine the extraction condition of the learning data. ..

In the condition setting information, an item that can specify the condition of the mechanical system 2 is selected as an item for which the condition is set. The status includes, for example, the operating mode (operating or non-operating) of the mechanical system 2 and the environmental status of the mechanical system 2 depending on the season. By using the parameter values of such items for the extraction of the learning data, it is possible to extract a data set having a similar situation where the mechanical system 2 is placed from the extraction source data.

The condition setting information is set in advance by the operator of the abnormality diagnosis device 1 or the like. The extraction unit 12 may hold a plurality of condition setting information. In this case, the extraction unit 12 may select the condition setting information used for extracting the learning data from the plurality of condition setting information according to the diagnosis target data. For example, the extraction unit 12 may select the condition setting information used for extracting the learning data according to the acquisition time zone of the diagnosis target data. The extraction unit 12 may select the condition setting information used for extracting the learning data according to the designation of the operator.

The creation unit 13 creates learning information corresponding to the diagnosis target data from the learning data received from the extraction unit 12. When the abnormality diagnosis device 1 makes a diagnosis using the MT method, the learning information is information indicating a unit space (a Mahalanobis space in the case of the MT method). The creation unit 13 outputs the learning information to the learning information storage unit 22 together with the number of extracted data sets included in the learning data, and stores the learning information and the number of extractions in the learning information storage unit 22.

The diagnosis unit 14 determines whether the diagnosis target data is abnormal (that is, whether the mechanical system 2 is abnormal) based on the learning information created by the creation unit 13 and the diagnosis threshold value. To do. Specifically, the diagnosis unit 14 reads the learning information and the number of extracted items from the learning information storage unit 22 and, regarding the diagnosis target data acquired from the mechanical system 2, is necessary for the diagnosis of the sample data based on the learning information. Calculate the numerical value (abnormality score). The abnormality score is an index indicating the degree of abnormality of the diagnosis target data (mechanical system 2). The larger the abnormality score, the higher the possibility that the diagnosis target data (mechanical system 2) is abnormal, and the smaller the abnormality score, the higher the possibility that the diagnosis target data (mechanical system 2) is normal. When the abnormality diagnosis device 1 makes a diagnosis using the MT method, the diagnosis unit 14 calculates the Mahalanobis distance as an abnormality degree score based on the diagnosis target data and the learning information.

The diagnosis unit 14 determines whether or not the diagnosis target data (mechanical system 2) is abnormal by comparing the abnormality score and the diagnosis threshold. The diagnostic threshold is a threshold corresponding to the number of extracted data sets extracted as learning data by the extraction unit 12. The diagnosis unit 14 selects a threshold value associated with the number of extracted cases as a diagnosis threshold value from a plurality of threshold values stored in the threshold value storage unit 23. The diagnostic unit 14 determines that the mechanical system 2 (diagnosis target data) is abnormal when the abnormality degree score is larger than the diagnosis threshold. The diagnosis unit 14 determines that the mechanical system 2 (diagnosis target data) is normal when the abnormality score is equal to or lower than the diagnosis threshold. The diagnosis unit 14 outputs a diagnosis result indicating whether the mechanical system 2 (diagnosis target data) is normal or abnormal to the output unit 15.

The output unit 15 outputs the diagnosis result received from the diagnosis unit 14. In the present embodiment, the output unit 15 outputs the diagnosis result to the external device 3. The output unit 15 may output the result of diagnosis by the diagnosis unit 14 and the information related to the learning information to the external device 3 in combination. The output unit 15 prepares a diagnosis result and related information corresponding to a predetermined output format, and outputs the information to the external device 3. The output unit 15 may output the diagnosis target data to the updating unit 16 together with the diagnosis result.

The update unit 16 updates the extraction source data stored in the accumulated data storage unit 21. The update unit 16 updates the extraction source data by acquiring one or more normal data sets and adding the normal data sets to the extraction source data. The update unit 16 may acquire, for example, a data set input by the operator of the abnormality diagnosis device 1 into the abnormality diagnosis device 1 using the input device as a normal data set. The updating unit 16 may sequentially acquire a plurality of data sets from the mechanical system 2, and acquire a data set to which a status flag indicating normality is attached as a normal data set among the plurality of data sets.

For example, the operator of the abnormality diagnosis device 1 inputs a period (normal period) in which the mechanical system 2 was normal to the abnormality diagnosis device 1 using the input device. The updating unit 16 sets a state flag so that the data set obtained during the normal period of the mechanical system 2 out of the plurality of data sets input to the abnormality diagnosis device 1 is normal. The update unit 16 may set the status flag according to the diagnosis result of the diagnosis unit 14.

The update unit 16 updates the extraction source data at the update timing. The update timing is the timing at which it is detected that a specific event has occurred. Examples of the specific event are that maintenance of the mechanical system 2 is completed, that a certain period of time has passed since the extraction source data was updated last time, and that the abnormality score of a predetermined number of consecutive diagnostic target data is greater than the diagnostic threshold. Is large, and the operator inputs an update instruction to the abnormality diagnosis device 1 using the input device. That is, the extraction source data may be updated automatically or manually.

The setting unit 17 calculates a plurality of thresholds from the extraction source data stored in the accumulated data storage unit 21 and sets the plurality of thresholds in the threshold table of the threshold storage unit 23. More specifically, the setting unit 17 reads the extraction source data stored in the accumulated data storage unit 21 and prepares partial extraction source data and a verification data group from a plurality of data sets included in the extraction source data. To do. The partial extraction source data is a subset of a plurality of data sets included in the extraction source data. The verification data group is a subset of a plurality of data sets included in the extraction source data and is different from the partial extraction source data. For example, the setting unit 17 arbitrarily (randomly) extracts the partial extraction source data and the verification data group from the extraction source data. The setting unit 17 may divide the extraction source data into partial extraction source data and a verification data group. It should be noted that the partial extraction source data and the verification data group do not have to completely match, may not overlap, or may partially overlap. The data sets included in the extraction source data are all normal data sets, and thus each data set (verification data) included in the verification data group is a normal data set.

The setting unit 17 uses the verification data included in the verification data group and the condition setting information to extract the threshold learning data from the partial extraction source data. The threshold learning data is learning data for threshold setting. The method for extracting the threshold learning data is the same as the method for extracting the learning data, and thus detailed description will be omitted. The condition setting information may be the same as or different from the condition setting information held by the extraction unit 12. The setting unit 17 creates threshold learning information from the threshold learning data. The threshold learning information is learning information for setting a threshold. The method for creating the threshold learning information is the same as the method for creating the learning information, and thus detailed description thereof will be omitted.

The setting unit 17 calculates the abnormality score of the verification data using the threshold learning information. The setting unit 17 sets a plurality of thresholds based on the abnormality score and the number of extracted data sets included in the threshold learning data.

Specifically, the setting unit 17 calculates the abnormality score of each verification data as described above, and extracts the number of data sets extracted as threshold learning data of the verification data and the abnormality degree of the verification data. A distribution Ma (see FIG. 3) indicating the relationship between the score and is acquired. The range of values that the abnormality score of the verification data can take differs depending on the number of extracted data sets included in the threshold learning data. More specifically, as shown in FIG. 3, the smaller the number of extractions, the larger the variation (range) of the values that the abnormality score can take. This is considered to be because the threshold learning information has a statistically meaningless average value when the number of extracted items is small. In addition, the number of extraction cases may be small because the state of the mechanical system 2 is in transition. Since abnormality diagnosis is performed using the threshold learning information corresponding to the data set at that time, it is considered that various abnormality degree scores are calculated.

The setting unit 17 calculates a curve showing the relationship between the number of extracted cases and the maximum value of the abnormality degree score by function fitting, for example, based on the distribution Ma. The setting unit 17 uses the calculated curve to set the maximum value of the abnormality degree score with respect to the number of extracted cases as the threshold value of the number of extracted cases. The setting unit 17 may divide the distribution Ma by the bin width of the number of extractions, and calculate the threshold value for each of the divided sections using a statistical amount such as a standard deviation. The threshold for each section may be calculated as the maximum value of the abnormality score included in the section, for example.

The setting unit 17 sets a plurality of thresholds in the threshold table at the setting timing. The setting unit 17 may set a plurality of thresholds before the mechanical system 2 is in operation, or may set a plurality of thresholds during the operation of the mechanical system 2. The setting timing may be, for example, that the extraction source data has been updated and that the operator has input a setting instruction to the abnormality diagnosis device 1 using the input device. That is, the threshold table may be set (reset) automatically or manually (reset).

Next, an abnormality diagnosis method performed by the abnormality diagnosis device 1 will be described with reference to FIG. FIG. 4 is a flowchart showing a series of processes of the abnormality diagnosis method performed by the abnormality diagnosis device of FIG. The series of processes shown in FIG. 4 is started, for example, every time the diagnosis target data is obtained in the mechanical system 2.

First, the acquisition unit 11 acquires diagnosis target data from the mechanical system 2 (step S01). Then, the acquisition unit 11 outputs the acquired diagnosis target data to the extraction unit 12 and the diagnosis unit 14.

Subsequently, when the extraction unit 12 receives the diagnosis target data from the acquisition unit 11, the extraction unit 12 selects the condition setting information corresponding to the diagnosis target data from the held plurality of condition setting information. Then, the extraction unit 12 determines the extraction condition of the learning data using the diagnosis target data and the condition setting information (step S02).

Subsequently, the extraction unit 12 reads the extraction source data stored in the accumulated data storage unit 21 from the accumulated data storage unit 21, and extracts learning data from the extraction source data based on the extraction condition (step S03). Specifically, the extraction unit 12 extracts, from the extraction source data, a data set that satisfies the extraction condition among the plurality of data sets included in the extraction source data. Then, the extraction unit 12 outputs the extracted learning data to the creation unit 13.

Subsequently, when the creation unit 13 receives the learning data from the extraction unit 12, the creation unit 13 creates learning information corresponding to the diagnosis target data from the learning data (step S04). Subsequently, the creating unit 13 determines whether the created learning information is appropriate (step S05). For example, if the number of data sets extracted was sufficient, but the extraction conditions were unsuitable, and the data sets used as learning data were extremely biased, the learning information was judged to be inappropriate. Can be done. In the creating unit 13, a criterion for determining whether or not the learning information is appropriate is set in advance. The creation unit 13 determines whether the learning information is appropriate based on the determination standard. When the creation unit 13 determines that the learning information is not appropriate (step S05; NO), the process thereafter is stopped. Then, a series of processes of the abnormality diagnosis method performed by the abnormality diagnosis device 1 is completed.

On the other hand, in step S05, when the creation unit 13 determines that the learning information is appropriate (step S05; YES), the learning information and the number of extractions are output to the learning information storage unit 22 to learn the learning information and the number of extractions. The information is stored in the information storage unit 22. Then, the creating unit 13 stores the learning information and the number of extractions in the learning information storage unit 22, and then notifies the diagnostic unit 14 that the learning information creating process is completed. Note that the creation unit 13 also notifies the diagnosis unit 14 that the processing has been stopped even when it is determined that the learning information is not appropriate.

Subsequently, when the diagnosis unit 14 receives a notification from the creation unit 13 that the learning information creation processing has been completed, the diagnosis unit 14 performs an abnormality diagnosis of the diagnosis target data based on the learning information (step S06). Specifically, the diagnosis unit 14 reads the learning information and the number of extracted items from the learning information storage unit 22 and calculates the abnormality score using the diagnosis target data and the learning information. Then, the diagnosis unit 14 acquires a threshold value associated with the number of extracted cases as a diagnosis threshold value from the threshold value table stored in the threshold value storage unit 23, and compares the abnormality degree score with the diagnosis threshold value to determine the diagnosis target. Determine if the data is abnormal. Then, the diagnosis unit 14 outputs a diagnosis result indicating whether the diagnosis target data is normal or abnormal to the output unit 15.

Subsequently, when the output unit 15 receives the diagnosis result from the diagnosis unit 14, the output unit 15 outputs the diagnosis result (step S07). In the present embodiment, the output unit 15 performs desired processing on the diagnosis result and then outputs the diagnosis result to the external device 3. At this time, the output unit 15 may delete the learning information and the number of extracted items stored in the learning information storage unit 22. The output unit 15 may output the diagnosis target data to the update unit 16 together with the diagnosis result. When the process is stopped, the diagnosis unit 14 outputs a diagnosis result indicating that the abnormality diagnosis is stopped to the output unit 15, and the output unit 15 performs the abnormality diagnosis on the external device 3 or the like. Notify that there was not. With the above, a series of processes of the abnormality diagnosis method performed by the abnormality diagnosis device 1 is completed.

Next, the update processing of the extraction source data will be described with reference to FIG. FIG. 5 is a flowchart showing the update processing of the extraction source data performed by the abnormality diagnosis device of FIG. The series of processes shown in FIG. 5 is started, for example, when a data set is transmitted from the mechanical system 2.

First, the updating unit 16 acquires a data set (step S11). The update unit 16 sequentially acquires a plurality of data sets from the mechanical system 2, for example. Then, the updating unit 16 sets a status flag in each of the acquired plurality of data sets (step S12). For example, the operator of the abnormality diagnosis device 1 inputs the normal period of the mechanical system 2 to the abnormality diagnosis device 1 using the input device. The updating unit 16 sets a state flag so that the data set obtained during the normal period of the mechanical system 2 out of the plurality of data sets input to the abnormality diagnosis device 1 is normal. The update unit 16 may set the status flag according to the diagnosis result of the diagnosis unit 14.

Subsequently, the update unit 16 determines whether it is the update timing (step S13). The update unit 16 determines that it is the update timing, for example, when a specific event such as maintenance of the mechanical system 2 occurs (step S13; YES). Then, the updating unit 16 updates the extraction source data (step S14). Specifically, the updating unit 16 acquires, as a normal data set, a data set to which a status flag indicating normal is attached, from the plurality of data sets acquired in step S11, and extracts the normal data set from the extraction source. The extraction source data is updated by adding to the data. Then, the updating unit 16 performs a series of processes again.

On the other hand, in step S13, when the update unit 16 determines that it is not the update timing (step S13; NO), the series of processes is performed again without updating the extraction source data.

In this way, the data set to which the status flag indicating normal is added is added to the extraction source data at each update timing, and the number of data sets included in the extraction source data increases.

Next, the threshold table setting process will be described with reference to FIG. FIG. 6 is a flowchart showing a threshold table setting process performed by the abnormality diagnosis device of FIG. The series of processes shown in FIG. 6 is started, for example, when the operator of the abnormality diagnosing device 1 inputs an instruction to set the threshold table to the abnormality diagnosing device 1 using the input device. This process is performed, for example, before the abnormality diagnosis device 1 makes an abnormality diagnosis.

First, the setting unit 17 reads the extraction source data stored in the accumulated data storage unit 21 and prepares the partial extraction source data and the verification data group (step S21).

Subsequently, the setting unit 17 selects (acquires) one piece of verification data from the plurality of pieces of verification data included in the verification data group (step S22). Then, the setting unit 17 selects the condition setting information corresponding to the verification data from the held plurality of condition setting information. Then, the setting unit 17 determines the extraction condition of the threshold learning data using the verification data and the condition setting information (step S23). The process of step S23 is similar to the process of step S02. That is, the process of step S23 is the same as the extraction condition determination process performed by the extraction unit 12.

Subsequently, the setting unit 17 extracts the threshold learning data from the partial extraction source data based on the extraction condition (step S24). The process of step S24 is similar to the process of step S03. That is, the process of step S24 is similar to the learning data extraction process performed by the extraction unit 12.

Subsequently, the setting unit 17 creates threshold learning information corresponding to the verification data from the threshold learning data (step S25). The process of step S25 is similar to the process of step S04. That is, the process of step S25 is the same as the learning information creation process performed by the creation unit 13.

Subsequently, the setting unit 17 calculates the abnormality degree score of the verification data using the verification data and the threshold learning information (step S26). For example, the setting unit 17 calculates the abnormality score by the MT method.

Then, the setting unit 17 determines whether all the verification data included in the verification data group have been selected (step S27). When the setting unit 17 determines that there is unselected verification data (step S27; NO), it selects one piece of verification data from the unselected verification data (step S22). Then, the setting unit 17 performs the processing of steps S23 to S26 on the selected verification data, and makes the determination of step S27 again.

On the other hand, when the setting unit 17 determines in step S27 that all the verification data have been selected (step S27; YES), the setting unit 17 calculates the threshold value (step S28). For example, the setting unit 17 creates, for each of the plurality of verification data, a distribution Ma (scatter diagram) indicating the relationship between the number of data sets extracted as the threshold learning data and the abnormality score of the verification data. To do. Here, the verification data group includes data sets in all normal states of the same mechanical system 2. Therefore, even if the threshold learning data with the same number of extractions are included, the data sets included in each are different from each other. Then, the setting unit 17 calculates a plurality of threshold values based on the distribution Ma. For example, the setting unit 17 divides the distribution Ma by the bin width of the number of extracted cases, and calculates the threshold value for each of the divided sections by using the statistical amount such as the standard deviation.

Subsequently, the setting unit 17 sets the calculated threshold value for each extracted number (section) in the threshold value table stored in the threshold value storage unit 23 (step S29). With the above, the threshold table setting process performed by the abnormality diagnosis device 1 is completed.

Next, an abnormality diagnosis program P for causing a computer to function as the abnormality diagnosis device 1 will be described with reference to FIG. 7. FIG. 7 is a diagram showing the structure of the abnormality diagnosis program.

As shown in FIG. 7, the abnormality diagnosis program P includes a main module P10, an acquisition module P11, an extraction module P12, a creation module P13, a diagnosis module P14, an output module P15, an update module P16, and a setting module P17. The main module P10 is a part that integrally controls processing related to abnormality diagnosis. The functions realized by executing the acquisition module P11, the extraction module P12, the creation module P13, the diagnosis module P14, the output module P15, the update module P16, and the setting module P17 are respectively the acquisition unit 11 and the extraction unit in the above embodiment. The functions of 12, the creating unit 13, the diagnosis unit 14, the output unit 15, the updating unit 16, and the setting unit 17 are the same.

The abnormality diagnosis program P is provided in a fixedly recorded state in a tangible recording medium such as a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), and a semiconductor memory. Good. The abnormality diagnosis program P may be provided as a data signal superimposed on a carrier wave via a communication network.

Next, the effects of the abnormality diagnosis device 1, the abnormality diagnosis method, the abnormality diagnosis program P, and the recording medium will be described with reference to FIGS. 8A and 8B. FIG. 8A is a diagram for explaining the diagnosis threshold value in the abnormality diagnosis device of FIG. 1. FIG. 8B is a diagram for explaining a diagnosis threshold in the abnormality diagnosis device according to the comparative example. The horizontal axes of (a) of FIG. 8 and (b) of FIG. 8 indicate the number of extractions of the data set extracted as the learning data. The vertical axis in each of FIGS. 8A and 8B represents an abnormality degree score. The abnormality diagnosis device according to the comparative example mainly differs from the abnormality diagnosis device 1 in that a constant diagnosis threshold is used.

The diagnosis threshold needs to be set so that the abnormality degree score of the normal diagnosis target data is below the diagnosis threshold and the abnormality degree score of the abnormal diagnosis target data is above the diagnosis threshold. In the abnormality diagnosing device 1 and the abnormality diagnosing device according to the comparative example, the extraction condition is determined using the diagnosis target data and the condition setting information for determining the extraction condition, and the plurality of data sets included in the extraction source data are determined. Of these, a group of data sets satisfying the extraction condition is extracted as learning data, and learning information used for the pattern recognition method is created from the learning data. When the diagnosis target data is diagnosed using the learning information created in this way, the range of values that the abnormality score can take varies depending on the number of data sets extracted, even if the diagnosis target data is normal. .. Specifically, the smaller the number of extracted cases, the larger the variation (range) of the values that the abnormality score can take.

As shown in (b) of FIG. 8, in the abnormality diagnosis device according to the comparative example, the diagnosis threshold Dth is set to a certain abnormality score. In this case, although the diagnostic target data located in the region R1 of the scatter diagram (distribution diagram) is normal, its abnormality degree score is larger than the diagnostic threshold Dth, so that it is diagnosed as abnormal. That is, when the number of extracted cases is small, “erroneous detection” occurs in which normal diagnosis target data is diagnosed as abnormal. In addition, the diagnosis target data located in the region R2 of the scatter diagram is diagnosed as normal because its abnormality degree score is smaller than the diagnosis threshold Dth, although it is abnormal. That is, when the number of extracted items is large, “undetected” occurs in which abnormal diagnosis target data is diagnosed as normal.

The abnormality diagnosis apparatus according to the comparative example may be configured not to perform abnormality diagnosis as a countermeasure against erroneous detection when the number of extracted cases is smaller than the predetermined number Nth. The number of cases Nth is a lower limit value of the number of cases to be extracted that can maintain the reliability of the diagnosis result (learning information). In this case, for example, if the external environment changes significantly, the number of extracted cases will decrease, so that abnormality diagnosis is not performed. Therefore, the diagnostic ability is reduced.

On the other hand, as shown in FIG. 8A, in the abnormality diagnosis device 1, the diagnosis threshold Dth is changed according to the number of extracted cases. Specifically, the diagnostic threshold value Dth is changed in accordance with the number of extracted data sets included in the learning data in consideration of the range of the abnormality score of the normal diagnosis target data. That is, it is determined whether or not the diagnosis target data is abnormal based on the learning information and the diagnosis threshold value according to the number of extractions of the data set extracted as the learning data.

More specifically, the diagnostic threshold Dth is set to a larger value as the number of extracted cases is smaller. As a result, when the number of extracted items is small, the diagnosis target data is diagnosed as normal even if the abnormality score is large to some extent, so that it is possible to reduce the occurrence of false detection. Further, even when the number of extracted cases is smaller than the number of cases Nth and the reliability of the diagnosis result (learning information) is low, the abnormality diagnosis can be performed. Furthermore, the diagnostic threshold Dth is set to a smaller value as the number of extracted data sets is larger. As a result, when the number of extracted cases is large, the diagnosis target data is diagnosed as abnormal even if the abnormality score is not very large, so that it is possible to reduce the number of undetected cases. As a result, the diagnostic accuracy can be improved.

For example, the diagnosis target data D1 is a data set at the time of abnormality with a small sign level. In the abnormality diagnosis device according to the comparative example, the diagnosis target data D1 is diagnosed as normal and undetected. On the other hand, in the abnormality diagnosis device 1, the diagnosis target data D1 is diagnosed as an abnormality. In this way, the abnormality diagnosis device 1 can detect even an abnormality having a small sign level.

The threshold storage unit 23 stores a threshold table that manages a plurality of thresholds that are associated with the number of extracted data sets. The diagnosis unit 14 selects a threshold value associated with the number of extracted cases as a diagnosis threshold value from a plurality of threshold values in the threshold value table. In this way, the diagnostic threshold value is set only by acquiring the threshold value associated with the number of extracted items from the threshold value storage unit 23, so that the setting of the diagnostic threshold value can be simplified.

The abnormality diagnosis apparatus 1 creates a distribution Ma indicating the range in which the abnormality score can be obtained for each number of extractions from the extraction source data. Specifically, partial extraction source data that is a subset of a plurality of data sets included in the extraction source data and a verification data group that is a subset of a plurality of data sets included in the extraction source data are prepared. .. Then, using the verification data included in the verification data group and the condition setting information, the threshold learning data is extracted from the partial extraction source data, the threshold learning information is created from the threshold learning data, and the threshold learning information is used. The abnormality score of the verification data is calculated. Each of the plurality of data sets included in the extraction source data is a data set when the mechanical system 2 is in a normal state. Therefore, by calculating the abnormality degree score using the extraction source data, it is possible to grasp the range that the abnormality degree score of the normal diagnosis target data can take. Specifically, the abnormality score calculated for each verification data, and the number of extractions of the data set included in the threshold learning data that is the source of the threshold learning information used when calculating the abnormality score, A distribution Ma indicating the relationship of is obtained. By using this distribution Ma, a plurality of threshold values respectively associated with the number of extracted data sets can be set based on the range in which the abnormality score of normal diagnosis target data can be set, so that the diagnosis threshold value is optimal. Can be converted.

The setting unit 17 may reset the threshold table in response to the extraction source data being updated by the updating unit 16. In this case, the threshold table is reset when the extraction source data is updated by adding the normal data set. As a result, the diagnostic threshold value can be selected from a plurality of threshold values that reflect the latest situation of the mechanical system 2, so that the diagnostic accuracy can be further improved.

Note that the abnormality diagnosis device, the abnormality diagnosis method, the abnormality diagnosis program, and the recording medium according to the present disclosure are not limited to the above embodiment.

For example, in the above embodiment, the abnormality diagnosis device 1 is composed of one device, but the abnormality diagnosis device 1 may be composed of a plurality of devices.

Instead of the processor 10, an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), an FPGA (Field Programmable Gate Array), or the like may be used.

The division of the functional blocks shown in FIG. 1 is an example, and the abnormality diagnosis device 1 may be divided into other functional blocks according to the functions thereof. In addition, each functional unit of the abnormality diagnosis device 1 may be further subdivided, and some functional units may be integrated into one functional unit.

Further, the normal state of the mechanical system 2 may change due to external factors. For example, when maintenance of the mechanical system 2 is performed, the normal state of the mechanical system 2 may change. An example of maintenance of the mechanical system 2 is replacement of a sensor provided in the mechanical system 2. Therefore, each data set stored in the extraction source data may further include a change flag indicating whether the data set is a data set before the normal state is changed or a data set after the normal state is changed. Good. The change flag indicates that the data set has changed to a normal state by default. When the maintenance of the mechanical system 2 is performed, the updating unit 16 sets the change flag of the data set having the time information indicating the time before the completion of the maintenance to the data set before the normal state changes. Set as shown. Then, the extraction unit 12 may refer to the change flag and extract the learning data from the data set after the normal state has changed among the plurality of data sets included in the extraction source data. Alternatively, the updating unit 16 may update the extraction source data by deleting the data set before the normal state changes from the extraction source data.

According to this configuration, since the learning data is extracted after excluding the data set before the normal state change, it is possible to reduce the time required to search for the data set that satisfies the extraction condition. Further, by deleting the data set before the normal state changes from the extraction source data, the number of data sets held in the accumulated data storage unit 21 can be reduced, and the capacity of the accumulated data storage unit 21 can be reduced. It becomes possible to do.

The operator of the abnormality diagnosis device 1 may be configured to be able to change the state flag of the data set stored in the accumulated data storage unit 21 using the input device. For example, the status flag may be changed to indicate that it is abnormal when it is determined that the detection is erroneous.

Although the diagnostic accuracy improves as the number of extracted data sets included in the learning data increases, the diagnostic accuracy hardly changes when the number of extracted data increases to some extent even if the number of extracted data increases. On the other hand, the larger the number of extractions, the longer the processing time. Therefore, when the number of data sets extracted exceeds the upper limit number (upper limit value), the extraction unit 12 sorts a group of data sets satisfying the extraction condition according to a predetermined sort criterion (sort criterion). (Sorting), and the upper limit number of data sets from the beginning may be extracted as learning data. The upper limit number is a number that can sufficiently explain the statistic, and is, for example, a test statistic. The upper limit number is set to about 1000, for example.

As the sorting standard, for example, a sequence closer to the standard time is used. Examples of the reference time point include when the mechanical system 2 is diagnosed and when the mechanical system 2 is shipped. When the diagnosis time is used as the reference time point, the extraction unit 12 sorts the group of data sets that satisfy the extraction condition in ascending order of the date (date and time) (that is, the order close to the diagnosis time of the diagnosis target data). When the shipment time of the mechanical system 2 is used as the reference time point, the extraction unit 12 sorts the group of data sets that satisfy the extraction condition in ascending order of date (date and time) (that is, from the shipment time of the mechanical system 2). To do.

As the sorting criterion, the order in which the difference from the criterion value is the smallest may be used. An example of the reference value is a parameter value of a certain item among a plurality of items included in the diagnosis target data. Specifically, the extraction unit 12 selects a group of data sets satisfying the extraction condition in the order of decreasing difference (absolute value of difference) from the parameter value included in the diagnosis target data for a certain item of the plurality of items. Sort. As such an item, for example, setting output is used. Also, an item indicating the external environment such as intake air temperature may be selected. As the sorting criterion, the order from the smallest abnormality score may be used.

In this case, the setting unit 17 may set a threshold value for each number of extracted items in the range of the number of extracted items up to the upper limit number. As a result, the number of thresholds held in the threshold table can be reduced, so that the capacity of the threshold storage unit 23 can be reduced. Further, the maximum number of data sets included in the learning data is limited to the upper limit number. Therefore, it is possible to suppress an increase in time required for creating learning information and the like. As a result, it is possible to suppress an increase in the time required for the diagnosis while performing the abnormality diagnosis according to the situation of the mechanical system 2.

In the above-described embodiment, the setting unit 17 prepares a verification data group including only a normal data set, but an abnormal verification data group may be prepared in addition to this verification data group. Each verification data included in the abnormality verification data group is a data set (abnormal data set) when the mechanical system 2 is in an abnormal state. That is, the abnormal data set includes parameter values of a plurality of items when the mechanical system 2 is in an abnormal state. The abnormality verification data group is input to the abnormality diagnosis device 1 by the operator of the abnormality diagnosis device 1, for example.

As shown in FIG. 9, the setting unit 17 calculates an abnormality degree score for each piece of verification data included in the abnormality verification data group in the same manner as the verification data group, and the distribution Mb of the abnormality degree scores for each number of extracted cases To create more. The setting unit 17 calculates a boundary line B that separates the distribution Ma and the distribution Mb using a known classification method (classifier or the like). The setting unit 17 sets, for example, the abnormality score for each number of extractions indicated by the boundary line B as a threshold for the number of extractions. In this case, the threshold value is set for each number of extractions in consideration of not only the range of the abnormality degree score of the normal diagnosis target data but also the range of the abnormality degree score of the abnormal diagnosis target data. Therefore, the diagnostic threshold is further optimized, and the diagnostic accuracy can be further improved.

1 Abnormality diagnosis device 2 Mechanical system (diagnosis target system)
3 external device 10 processor 11 acquisition unit 12 extraction unit 13 creation unit 14 diagnosis unit 15 output unit 16 update unit 17 setting unit 20 storage device 21 accumulated data storage unit (first storage unit)
22 learning information storage unit 23 threshold value storage unit (second storage unit)
P Abnormality diagnosis program

Claims (9)

An abnormality diagnosis device for performing an abnormality diagnosis of a diagnosis target system using a pattern recognition method,
An acquisition unit that acquires diagnosis target data including parameter values of a plurality of items from the diagnosis target system,
A first storage unit for storing extraction source data including a plurality of data sets;
An extraction unit that determines the extraction condition using the diagnosis target data and the condition setting information for determining the extraction condition, and extracts learning data from the extraction source data,
A creation unit that creates learning information used in the pattern recognition method from the learning data,
Based on the learning information and a diagnosis threshold value according to the number of extracted data sets extracted as the learning data by the extraction unit, a diagnosis unit that determines whether or not the diagnosis target data is abnormal,
Equipped with
Each of the plurality of data sets includes parameter values of the plurality of items when the system to be diagnosed is in a normal state,
The abnormality diagnosis device, wherein the extraction unit extracts, as the learning data, a group of data sets satisfying the extraction condition from the plurality of data sets included in the extraction source data. The abnormality diagnosis device according to claim 1, wherein the diagnosis threshold is set to a larger value as the number of extracted cases is smaller. A second storage unit for storing a plurality of threshold values respectively associated with the number of data sets,
The abnormality diagnosis device according to claim 1, wherein the diagnosis unit selects, from the plurality of threshold values, a threshold value associated with the extracted number as the diagnosis threshold value. Further comprising a setting unit for setting the plurality of thresholds,
The setting unit is a partial extraction source data that is a first subset of the plurality of data sets included in the extraction source data, and a verification that is a second subset of the plurality of data sets included in the extraction source data. Prepare the data group and
The setting unit uses the verification data included in the verification data group and the condition setting information to extract threshold learning data, which is learning data for threshold setting, from the partial extraction source data, and performs the threshold learning. Create threshold learning information that is learning information for threshold setting from the data,
The setting unit calculates an abnormality score of the verification data using the threshold learning information, and sets the plurality of thresholds based on the abnormality score and the number of data sets included in the threshold learning data. The abnormality diagnosis device according to claim 3, wherein The setting unit further prepares an abnormality verification data group including a plurality of abnormal data sets,
Based on the first distribution acquired based on the verification data group and the second distribution acquired based on the abnormality verification data group, the plurality of threshold values are set,
Each of the plurality of abnormal data sets, including the parameter value of the plurality of items when the system to be diagnosed is in an abnormal state,
The abnormality diagnosis device according to claim 4, wherein each of the first distribution and the second distribution indicates a possible range of the abnormality degree score for each of the extracted cases. Further comprising an updating unit for updating the extraction source data stored in the first storage unit,
The update unit updates the extraction source data by adding a normal data set to the extraction source data,
The abnormality diagnosis device according to claim 4 or 5, wherein the setting unit resets the plurality of thresholds in response to the extraction source data being updated. An abnormality diagnosis method executed by an abnormality diagnosis device for performing abnormality diagnosis of a system to be diagnosed using a pattern recognition method,
Acquiring diagnostic target data including parameter values of a plurality of items from the diagnostic target system,
Determining the extraction condition using the diagnosis target data and condition setting information for determining the extraction condition,
Extracting the training data from the source data containing multiple datasets,
Creating learning information used in the pattern recognition method from the learning data,
Based on the learning information and a diagnosis threshold value according to the number of extracted data sets extracted as the learning data, a step of determining whether or not the diagnosis target data is abnormal,
Equipped with
Each of the plurality of data sets includes parameter values of the plurality of items when the system to be diagnosed is in a normal state,
The abnormality diagnosis method, wherein in the extracting step, a group of data sets satisfying the extraction condition among the plurality of data sets included in the extraction source data is extracted as the learning data. An abnormality diagnosis program for causing a computer to execute an abnormality diagnosis related to a diagnosis target system using a pattern recognition method,
Acquiring diagnostic target data including parameter values of a plurality of items from the diagnostic target system,
Determining the extraction condition using the diagnosis target data and condition setting information for determining the extraction condition,
Extracting the training data from the source data containing multiple datasets,
Creating learning information used in the pattern recognition method from the learning data,
Based on the learning information and a diagnosis threshold value according to the number of extracted data sets extracted as the learning data, a step of determining whether or not the diagnosis target data is abnormal,
To the computer,
Each of the plurality of data sets includes parameter values of the plurality of items when the system to be diagnosed is in a normal state,
In the extracting step, of the plurality of data sets included in the extraction source data, a group of data sets satisfying the extraction condition is extracted as the learning data,
Abnormality diagnosis program. A computer-readable recording medium having recorded therein an abnormality diagnosis program for causing a computer to perform abnormality diagnosis related to a diagnosis target system using a pattern recognition method,
Acquiring diagnostic target data including parameter values of a plurality of items from the diagnostic target system,
Determining the extraction condition using the diagnosis target data and condition setting information for determining the extraction condition,
Extracting the training data from the source data containing multiple datasets,
Creating learning information used in the pattern recognition method from the learning data,
Based on the learning information and a diagnosis threshold value according to the number of extracted data sets extracted as the learning data, a step of determining whether or not the diagnosis target data is abnormal,
To the computer,
Each of the plurality of data sets includes parameter values of the plurality of items when the system to be diagnosed is in a normal state,
In the extracting step, of the plurality of data sets included in the extraction source data, a group of data sets satisfying the extraction condition is extracted as the learning data,
A recording medium recording an abnormality diagnosis program.