JP2016091417A - Abnormality diagnostic method, abnormality diagnostic apparatus and abnormality diagnostic program - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce dependence on a selector and easily select unit data.SOLUTION: An abnormality diagnostic apparatus 100 diagnoses whether a diagnosis object is abnormal on the basis of a dataset having parameter values for a plurality of items related to the diagnosis object while using a diagnostic scheme on the premise that the parameter values follow a normal distribution. The abnormality diagnostic apparatus 100 comprises a unit data selection part 21 that selects unit data D2 on the basis of which a diagnosis is conducted from selection data D1 including a plurality of datasets. The unit data selection part 21 includes a first calculation section 21a calculating a first range based on a standard deviation of the parameter values for each of the items in the datasets included in the selection data D1; and a selection section 21b selecting the datasets the parameter values of which are present in the first range for all the items out of the datasets included in the selection data D1.SELECTED DRAWING: Figure 1

Description

  The present invention relates to an abnormality diagnosis method, an abnormality diagnosis apparatus, and an abnormality diagnosis program.

  As a technique related to a conventional abnormality diagnosis method, for example, Patent Document 1 describes a plant monitoring method for monitoring an operation state of a plant having a plurality of evaluation items. In the method described in Patent Literature 1, a state quantity file that is a collection of state quantities (data sets) that is a collection of state quantities (parameter values) for each of a plurality of evaluation items is stored. Extract the bundle. Then, based on the data (unit data) stored in the unit space file, the Mahalanobis distance of the extracted bundle of state quantities is obtained, and the operation of the plant is determined depending on whether or not the Mahalanobis distance is within a predetermined threshold. Determine whether the state is normal.

JP 2012-252556 A

  In the above prior art, there is no clear method for selecting unit data as a reference when making a diagnosis. Usually, as unit data, a data set in a period in which no abnormality has occurred in the diagnosis target, a data set in which the sensor values (parameter values) of a plurality of sensors in the diagnosis target are in a normal range, and the like are selected. However, in such unit data selection, there is a possibility that the selection work depends on the experience and feeling of the selector. In addition, for example, if there are hundreds of sensors, it may be necessary to determine whether or not the sensor values of these sensors are in the normal range. is there.

  Therefore, an object of the present invention is to provide an abnormality diagnosis method, an abnormality diagnosis apparatus, and an abnormality diagnosis program that can easily select unit data while reducing the dependence on a selector.

  The abnormality diagnosis method according to the present invention is based on a data set having a plurality of parameter values related to a diagnosis target, and uses a diagnosis method based on the assumption that the parameter values follow a normal distribution. A method comprising a data selection step of selecting unit data serving as a reference when performing diagnosis from selection data including a plurality of data sets, wherein the data selection step includes a plurality of data sets included in the selection data. In the step, the first range based on the standard deviation of the parameter value is calculated for each of the plurality of items, and the parameter values are present in the first range in all of the plurality of items among the plurality of data sets included in the selection data. Selecting a data set as unit data.

  The abnormality diagnosis apparatus according to the present invention is based on a data set having parameter values of a plurality of items related to a diagnosis target, and uses a diagnosis method based on the assumption that the parameter values follow a normal distribution. The apparatus includes a unit data selection unit that selects unit data serving as a reference when performing diagnosis from selection data including a plurality of data sets, and the unit data selection unit includes a plurality of data included in the selection data. In the data set, for each of a plurality of items, the first calculation unit that calculates the first range based on the standard deviation of the parameter value, and the parameter value is the first in all of the plurality of items among the plurality of data sets included in the selection data. And a selection unit that selects a data set existing within the range as unit data.

  In such an abnormality diagnosis method and an abnormality diagnosis apparatus, among the plurality of data sets included in the selection data, a data set in which parameter values exist in the first range in all of the plurality of items (that is, out of the first range). Other than the data set having the parameter value) is selected as the unit data. Therefore, unit data can be selected from the selection data without going through the judgment of the selector and without determining whether each parameter value is in the normal range. Therefore, it is possible to easily select unit data while reducing the dependency on the selector.

  The abnormality diagnosis method according to the present invention further includes a data excluding process for excluding a part of the selection data, and the data excluding process includes a parameter value for each item in a plurality of data sets included in the selection data. A step of calculating the second range based on the standard deviation of the degree and the parameter value, and when the kurtosis is equal to or greater than a predetermined value, the plurality of data sets included in the selection data have parameter values that are out of the second range And excluding the data set.

  The abnormality diagnosis apparatus according to the present invention further includes a selection data exclusion unit that excludes a part of the selection data, and the selection data exclusion unit is divided into a plurality of items in a plurality of data sets included in the selection data. A second calculation unit that calculates a kurtosis of the parameter value and a second range based on the standard deviation of the parameter value; and a second range from a plurality of data sets included in the selection data when the kurtosis is equal to or greater than a predetermined value And an exclusion unit that excludes a data set having a parameter value that is out of the range.

  In such an abnormality diagnosis method and abnormality diagnosis apparatus, when the kurtosis is equal to or greater than a predetermined value, a data set having a parameter value (so-called outlier) that is out of the second range is excluded from the selection data. Thereby, the normality of the data for selection can be improved.

  The abnormality diagnosis method according to the present invention diagnoses whether a diagnosis target is abnormal or normal based on a data acquisition step of acquiring diagnostic data, which is a diagnostic data set from a diagnosis target, and unit data and diagnostic data. A diagnosis process and a data update process for replicating unit data as selection data and adding diagnosis data to the selection data when the diagnosis target is diagnosed as normal in the diagnosis process may be provided.

  An abnormality diagnosis apparatus according to the present invention is based on a diagnosis data acquisition unit that acquires diagnosis data that is a data set for diagnosis from a diagnosis object, and based on unit data and diagnosis data, whether the diagnosis object is abnormal or normal An abnormality diagnosis unit to diagnose, a selection data update unit that duplicates unit data as selection data, and adds diagnosis data to the selection data when the diagnosis target is diagnosed as normal by the abnormality diagnosis unit; May be provided.

  In such an abnormality diagnosis method and abnormality diagnosis device, it is possible to update the selection data every time an abnormality diagnosis is made.

  In the abnormality diagnosis method and the abnormality diagnosis apparatus according to the present invention, the diagnosis method may be a Mahalanobis Taguchi method, and the unit data may constitute a unit space serving as a reference when calculating the Mahalanobis distance. In this case, in the abnormality diagnosis by the Mahalanobis Taguchi method, the unit data constituting the unit space used as a reference when calculating the Mahalanobis distance can be easily selected with less dependence on the selector.

  The abnormality diagnosis program according to the present invention causes a computer to execute the abnormality diagnosis method. Also in this abnormality diagnosis program, the same effects as the effects in the abnormality diagnosis method are exhibited.

  ADVANTAGE OF THE INVENTION According to this invention, it becomes possible to provide the abnormality diagnosis method, abnormality diagnosis apparatus, and abnormality diagnosis program which can select unit data easily, reducing the dependence by a selector.

It is a block diagram which shows the structure of the abnormality diagnosis apparatus which concerns on one Embodiment. It is a figure which shows the concept of MT method. It is a figure explaining an example of the data for selection. It is a graph which shows the example of distribution of a parameter value. It is a flowchart which shows an abnormality diagnosis process. It is a flowchart which shows a unit data determination process. (A) is a flowchart which shows the data exclusion process for selection. (B) is a flowchart showing unit data selection processing. It is a flowchart which shows the data update process for selection. It is a figure for demonstrating an abnormality diagnosis program.

  Hereinafter, embodiments of the present invention will be described in detail with reference to the accompanying drawings. In the following description, the same or equivalent elements will be denoted by the same reference numerals, and redundant description will be omitted.

  FIG. 1 is a block diagram showing the configuration of the abnormality diagnosis apparatus according to this embodiment. FIG. 2 is a diagram illustrating the concept of the MT method. FIG. 3 is a diagram illustrating an example of selection data. FIG. 4 is a graph showing a distribution example of parameter values of one item in a plurality of data sets included in the selection data.

  As shown in FIG. 1, the abnormality diagnosis device 100 diagnoses an abnormality of a diagnosis target based on a data set having a plurality of parameter values related to the diagnosis target. The object to be diagnosed is not particularly limited, and examples thereof include various plants including a gas turbine power plant and a mechanical system having a plurality of devices. In the present embodiment, a power generation plant including a gas turbine is applied as a diagnosis target.

  The data set is a data group in which parameter values of a plurality of items are collected, and is also referred to as sample data. The parameter values of a plurality of items are data values related to a plurality of types of feature amounts (variables). The parameter value items include, for example, control values, command values and response values of a plurality of types of devices (valves, pumps, etc.) included in the diagnosis target, and sensor values of a plurality of types of sensors provided in the diagnosis target. Is mentioned. The number of parameter value items is not particularly limited, but is several hundred or more here.

  In the abnormality diagnosis in the present embodiment, a diagnosis method is used on the assumption that each parameter value of the data set follows a normal distribution. In the diagnostic method, unit data that is normal data is used as a reference for diagnosis, and the content of the unit data affects the diagnostic accuracy. An example of such a diagnostic technique is the MT (Mahalanobis Taguchi) method. In addition to the above, as a diagnostic method on the assumption that the parameter value follows a normal distribution, for example, it may be used for regression analysis or the like.

  In the present embodiment, the MT method is adopted as a diagnostic method. The MT method is one of pattern recognition methods proposed in quality engineering, and diagnoses whether or not it is the same as usual. In the MT method, even a data set having several hundred or more feature amounts (items) can be handled as one judgment index. As shown in FIG. 2, in the abnormality diagnosis by the MT method, a unit space is generated from unit data, and the Mahalanobis distance (hereinafter referred to as “MD value”) from the diagnostic data to be diagnosed using this unit space as a reference for diagnosis. To determine whether it is abnormal or normal.

  When the MD value is smaller than the threshold value, it is diagnosed as normal because it is close to the unit space, that is, in a normal state. On the other hand, when the MD value is equal to or greater than the threshold value, an abnormality is diagnosed as being in a state far from the unit space, that is, a state away from the normal state. The unit space is a normal group (uniform group) that is homogeneous for the purpose and has homogeneity. This unit space can also be said to be “steady and normal state”, “frequent state” or “average state”.

  As shown in FIG. 1, the abnormality diagnosis apparatus 100 includes a storage device 10 and a CPU (Central Processing Unit) 20. The storage device 10 stores selection data D1 and unit data D2. Examples of the storage device 10 include a recording medium such as a ROM, a flexible disk, a CD-ROM, a DVD, or a USB memory, or a semiconductor memory.

  The selection data D1 is data used for selection of the unit data D2, and includes a plurality of data sets. The selection data D1 here is formed by a plurality of data sets that have been input and accumulated in the past. In the example shown in FIG. 3, the selection data D1 is formed by a plurality of data sets d that are input from the plant and accumulated every predetermined time. The selection data D1 may be formed without depending on the data set input from the plant, or may be formed by external input separately.

  The number of data sets d included in the selection data D1 is not particularly limited and can be set as appropriate. The selection data D1 shown in the figure includes a data set d for each predetermined time, but it is only necessary to include a plurality of data sets. Returning to FIG. 1, the unit data D2 is data used as a reference when making a diagnosis, and constitutes a unit space. The unit data D2 includes a plurality of normal data sets used for generating a unit space.

  The CPU 20 includes a unit data selection unit 21, a selection data exclusion unit 22, a diagnosis data acquisition unit 23, an abnormality diagnosis unit 24, and a selection data update unit 25 as functional components. The unit data selection unit 21 acquires selection data D1 from the storage device 10, and selects unit data D2 from the selection data D1. The unit data selection unit 21 includes a first calculation unit 21a, a selection unit 21b, and a correction unit 21c.

The first calculation unit 21a calculates a statistic including an average value Av _{1 of} parameter values, a standard deviation σ ₁ and a first range R1 for each of a plurality of items in a plurality of data sets included in the selection data D1. The first range R1 is a range specified based on the standard deviation sigma _1. Specifically, the first range R1 is a range based on an interval m · σ ₁ that is a multiplier of a specified value m and a standard deviation σ ₁ . More specifically, the first range R1 is an average value Av ₁ of the center that the average value Av ₁ interval m · sigma ₁ spreads the positive and negative side, respectively from confidence intervals _{(= Av 1 ± m · σ} 1) ( (See FIG. 4).

The designated value m is a preset value stored in advance. The designated value m may be designated by a selector (user) via an input device (not shown), or may be designated as a fixed value. The designated value m may be uniformly set to the same value for all items of the parameter value, or may be set to a different value for at least some items. The designated value m may be a different value for each item, for example, according to normal distribution information for each item. In addition, the designated value m may be corrected in a timely manner within the variation range of each item in accordance with the actual driving situation to be diagnosed. The designated value m here is set to “2” which is a positive real value in all items, and as a result, the range of “Av ₁ ± 2 · σ ₁ ” is the first in all of the plurality of items. The range is R1.

  The selection unit 21b includes, among the plurality of data sets included in the selection data D1, a data set in which parameter values exist in the first range R1 in all of the plurality of items (that is, parameter values that are out of the first range R1). A data set other than the data set) is selected as the unit data D2. In other words, when all of the plurality of parameter values in the data set are within the first range R1, the data set is selected as the unit data D2, and even at least one of the plurality of parameter values in the data set is the first range R1. If it is outside, the data set is not selected as unit data D2. Within the first range R1, the selection unit 21b outputs the selected unit data D2 to the storage device 10.

  The correction unit 21c corrects the unit data D2 selected by the selection unit 21b and output to the storage device 10 according to the operation input by the user input via the data correction unit 30. The data correction unit 30 includes an input unit 30a such as a keyboard and operation buttons, and a display unit 30b such as a display. In the data correction unit 30, the unit data D2 selected by the selection unit 21b and output to the storage device 10 is displayed on the display unit 30b, and an operation input is made by the user via the input unit 30a to correct the unit data D2. If so, the operation input is output to the correction unit 21c.

  If the unit space has not yet been generated from the start of diagnosis (when the unit space is generated in the abnormality diagnosis unit 24, the selection data excluding unit 22 is the first generation), the selection data exclusion unit 22 selects the selection data D1 from the storage device 10. And a part of the selection data D1 is excluded. The selection data excluding unit 22 includes a second calculating unit 22a and an excluding unit 22b.

The second calculation unit 22a includes, in a plurality of data sets included in the selection data D1, a statistic including an average value Av ₂ , a standard deviation σ ₂ , a kurtosis γ, and a second range R2 for each of the plurality of items. Is calculated. The kurtosis γ represents how much the parameter value distribution is concentrated around the average value Av ₂ . The kurtosis γ tends to show a large value when an outlier, which is a value greatly deviating from other values, is included as a parameter value. The kurtosis γ can be obtained by the following equation.
γ = ((a ₁ −Av ₂ ) ² + (a ₂ −Av ₂ ) ²
+ ... (a _n −Av ₂ ) ² ) / (n · σ ₂ ² )
a _{1 to} a _n : Parameter value

The second range R2 is a specified range based on the standard deviation sigma _2. Specifically, the second range R2 is a range based on a section k · σ ₂ that is a multiplier of the specified value k and the standard deviation σ ₂ . More specifically, the second range R2 is the average value Av ₂ was mainly the average value Av ₂ interval k · sigma ₂ spreads in the positive and negative side, respectively from confidence intervals _{(= Av 2 ± k · σ} 2).

The designated value k is a designated value stored in advance. The designated value k may be designated by the user via an input device (not shown) or may be designated as a fixed value. The designated value k may be set to the same value for all items of the parameter value, or may be set to a different value for at least some of the items. The designated value k may be a different value for each item according to, for example, normal distribution information for each item. Further, the designated value k may be corrected in a timely manner within the variation range of each item in accordance with the actual driving situation to be diagnosed. Here, the designated value k is set to “3” which is a positive real value larger than the designated value m for all items of the parameter value, and as a result, “Av ₂ ± 3” is set for all of the plurality of items. The range of “σ ₂ ” is the second range R2.

  When the kurtosis γ calculated by the second calculation unit 22a is equal to or greater than a predetermined value, the excluding unit 22b determines that the selection data D1 includes an outlier, and deviates from the parameter value outside the second range R2. The data set having the outlier is excluded from the selection data D1. In other words, when all of the plurality of parameter values in the data set are within the second range R2, the data set is not excluded from the selection data D1, while at least one of the plurality of parameter values in the data set is the second value. If it is outside the range R2, the data set is excluded from the selection data D1. The excluding unit 22b outputs the selection data D1 excluding the data set having an outlier to the storage device 10. The predetermined value is a value generally or empirically set in advance when determining the presence or absence of an outlier. The predetermined value is, for example, a value of 5 or more, and is a value of 10 or more in the case of more reliable determination.

  The diagnostic data acquisition unit 23 acquires diagnostic data, which is a diagnostic data set, from the diagnosis target. The diagnostic data acquisition unit 23 outputs the acquired diagnostic data to the abnormality diagnosis unit 24 and the selection data update unit 25.

  The abnormality diagnosis unit 24 acquires the unit data D2 from the storage device 10, acquires the diagnosis data from the diagnosis data acquisition unit 23, and the diagnosis target is abnormal based on the acquired unit data D2 and the diagnosis data. Diagnose whether it is normal.

  Specifically, the abnormality diagnosis unit 24 executes the following process. That is, the unit data D2 is acquired from the storage device 10, and a unit space is generated from the unit data D2. The diagnostic data is acquired from the diagnostic data acquisition unit 23, and the MD value is calculated using the generated unit space. When the MD value is smaller than the threshold value, the state of the plant is diagnosed as normal. On the other hand, when the MD value is equal to or larger than the threshold value, the diagnosis target state is diagnosed as abnormal. The threshold value is a value set and stored in advance. In addition, a well-known method can be used as a method of abnormality diagnosis based on such unit data D2 and diagnostic data.

  The abnormality diagnosis unit 24 outputs the diagnosis result to the selection data update unit 25. In addition, the abnormality diagnosis unit 24 outputs and notifies the diagnosis result to a notification device (not shown) such as an external display or a speaker.

  When the unit space is generated at least once from the start of diagnosis (when the unit space is generated by the abnormality diagnosis unit 24, the selection data update unit 25 generates data other than the first time). Update. Specifically, the unit data D2 is acquired from the storage device 10, and this unit data D2 is copied as selection data D1. When the abnormality diagnosis unit 24 determines that the diagnosis target is normal, the diagnosis data is added to the duplicated selection data D1 and output to the storage device 10, while the abnormality diagnosis unit 24 selects the diagnosis target. If it is not diagnosed as normal, the selection data D1 that is duplicated is output to the storage device 10 as it is.

  Next, an abnormality diagnosis method performed by the abnormality diagnosis apparatus 100 will be described.

  FIG. 5 is a flowchart showing the abnormality diagnosis process. As shown in FIG. 5, in the abnormality diagnosis apparatus 100, the CPU 20 repeatedly performs the next abnormality diagnosis process at a predetermined time period, and diagnoses the abnormality to be diagnosed at a predetermined time period. That is, first, unit data determination processing is executed to determine unit data D2, and the unit diagnosis D24 is acquired by the abnormality diagnosis unit 24 (S1). The diagnosis data output from the diagnosis target is acquired by the abnormality diagnosis unit 24 via the diagnosis data acquisition unit 23 (S2: data acquisition step).

  Subsequently, the abnormality diagnosis unit 24 generates a unit space based on the unit data D2 (S3). Specifically, in each of the plurality of items of the unit data D2, an average value and a standard deviation are obtained, and a normalized value is obtained from the parameter value, the average value, and the standard deviation, and a correlation matrix is generated. A unit space is generated by obtaining an inverse matrix of the generated correlation matrix. Then, the abnormality diagnosis unit 24 calculates the MD value based on the diagnostic data using the unit space (S4).

  Subsequently, it is determined whether or not the calculated MD value is smaller than a threshold value (S5). When the MD value is smaller than the threshold value, the diagnosis data is normal data, and it is diagnosed that the state of the diagnosis target is normal (S6). On the other hand, when the calculated MD value is greater than or equal to the threshold value, the diagnosis data is abnormal data, and the diagnosis target state is diagnosed as abnormal (S7). In addition, the process of said S3-said S7 comprises a diagnostic process.

  Here, in the abnormality diagnosis apparatus 100 of the present embodiment, as described above, when diagnosing abnormality of the diagnosis target, unit data determination processing is executed before generation of the unit space. Hereinafter, the unit data determination process will be described in detail.

  FIG. 6 is a flowchart showing the unit data determination process. As shown in FIG. 6, in the unit data determination process, it is first determined whether or not the unit space is generated for the first time (S11). Here, for example, an initial determination flag is provided in the execution of the process of S3, and the determination can be made based on whether or not the initial determination flag is set.

  In the case of YES in S11, the selection data exclusion unit 22 executes selection data exclusion processing for excluding outliers from the selection data D1 (S12: data exclusion process). Subsequently, the unit data selection unit 21 executes unit data selection processing for selecting the unit data D2 from the selection data D1 (S13: data selection step).

  In the case of NO in S11, the selection data update unit 25 executes selection data update processing for updating the selection data D1 (S14: data update process). Subsequently, similarly to S14, the unit data selection unit 21 executes unit data selection processing for selecting the unit data D2 from the selection data D1 (S15: data selection step).

  After S13 or S15, information on the selected unit data D2 is displayed on the display unit 30b of the data correction unit 30. Then, for example, when an operation input is made by the selector through the input unit 30a of the data correction unit 30 based on, for example, the distribution of the unit data D2 by item or past experience, the unit data D2 is corrected according to the operation input. (S16). Thus, the unit data D2 is determined, and this unit data D2 is output to the storage device 10 and stored.

FIG. 7A is a flowchart showing the selection data excluding process. As shown in FIG. 7A, in the selection data excluding process, the selection data D1 is acquired from the storage device 10, and the average of the parameter values is obtained for a plurality of items of a plurality of data sets included in the selection data D1. Statistics including value Av ₂ , standard deviation σ ₂ , kurtosis γ, and second range R 2 are calculated (S 21). Based on whether or not the kurtosis γ is greater than or equal to a predetermined value in any item, it is determined whether or not there is an outlier in the parameter values of a plurality of data sets included in the selection data D1 (S22).

  If the kurtosis γ is not greater than or equal to the predetermined value, it is determined that there are no outliers in the parameter values of the plurality of data sets included in the selection data D1, and the process is terminated. On the other hand, when the kurtosis γ is equal to or larger than the predetermined value, the data set having an outlier is excluded from the selection data D1 (S23). Specifically, a parameter value out of the second range R2 is regarded as an outlier, and a data set having the outlier among the data sets included in the selection data D1 is excluded from the selection data D1. After the outlier is excluded, the statistic is calculated again (S21), it is determined whether or not the outlier exists based on the kurtosis γ (S22), and the data set having the outlier is selected as the selection data D1. Is repeatedly performed until there is no outlier (the determination process in S22 is NO). Then, the selection data D1 is output to the storage device 10.

FIG. 7B is a flowchart showing unit data selection processing. As shown in FIG. 7B, in the unit data selection process, the selection data D1 is acquired from the storage device 10, and the average value of the parameter values for each of the plurality of items of the plurality of data sets included in the selection data D1. A statistic including Av ₁ , standard deviation σ ₁ , and first range R1 is calculated (S31). Among the plurality of data sets included in the selection data D1, the data sets in which the parameter values exist in the first range R1 in all of the plurality of items (that is, other than the data sets having parameter values outside the first range R1). Is selected as unit data D2 (S32).

  FIG. 8 is a flowchart showing the selection data update process. As shown in FIG. 8, in the selection data update process, the unit data D2 is acquired from the storage device 10, and this unit data D2 is duplicated as the selection data D1 (S41). Subsequently, it is determined whether or not the diagnosis result by the abnormality diagnosis unit 24 is normal (whether the determination process of S5 is YES) (S42). When the diagnosis result is normal, the diagnosis data (diagnosis data acquired in S2) used for the diagnosis is added to the selection data D1, and the selection data D1 is output to the storage device 10 (S43). . On the other hand, when the diagnosis result is not normal, the selection data D1 is output to the storage device 10 as it is without adding the diagnosis data to the selection data D1.

  Next, an abnormality diagnosis program that causes the CPU (computer) 20 to execute the above-described series of abnormality diagnosis processes will be described.

  FIG. 9 is a diagram for explaining the abnormality diagnosis program. As shown in FIG. 9, the abnormality diagnosis program P is inserted into the CPU 20 and accessed, and is stored in the program storage area 12 formed in the storage device 10. The abnormality diagnosis program P includes a unit data selection module 51, a selection data exclusion module 52, an abnormality diagnosis module 53, a diagnosis data acquisition module 54, and a selection data update module 55.

  The functions realized by the unit data selection module 51 are the same as the functions of the unit data selection unit 21 described above. The function realized by the selection data exclusion module 52 is the same as the function of the selection data exclusion unit 22 described above. The function realized by the abnormality diagnosis module 53 is the same as the function of the abnormality diagnosis unit 24 described above. The function realized by the diagnostic data acquisition module 54 is the same as the function of the diagnostic data acquisition unit 23 described above. The function realized by the selection data update module 55 is the same as the function of the selection data update unit 25 described above.

As described above, in the abnormality diagnosis device 100, the abnormality diagnosis method, and the abnormality diagnosis program P according to the present embodiment, the _first based on the standard deviation σ ₁ of the parameter value for each of a plurality of items in a plurality of data sets included in the selection data D1. One range R1 is calculated. Then, among the plurality of data sets included in the selection data D1, a data set having parameter values within the first range R1 in all of the plurality of items is selected as the unit data D2.

  This eliminates the need for the selector (person) to select the unit data D2, reduces the judgment intervention by the selector, and determines whether each parameter value is in the normal range, Unit data D2 can be selected. Therefore, the selection of the unit data D2 can be automatically formulated to a selectable level, the selection can be stabilized, and the unit data D2 can be easily selected with less dependence on the selector. Become. As a result, there is a problem that it depends on human experience and is troublesome (for example, depending on the selected person, abnormal data is mixed with the unit data D2, or if there are many items, it is difficult to check the parameter value for each item. Etc.) can be eliminated.

In the present embodiment, in a plurality of data sets included in the selection data D1, the kurtosis γ of the parameter value and the second range R2 based on the standard deviation σ ₂ of the parameter value are calculated for each of a plurality of items. When the kurtosis γ is equal to or greater than a predetermined value, a data set having a parameter value outside the second range R2 is excluded from a plurality of data sets included in the selection data D1. Thereby, a data set having an outlier can be excluded from the selection data D1, and the normality of the selection data D1 can be improved.

In particular, in the present embodiment, in each distribution of parameter values of several hundred or more items included in the selection data D1, those that fall within the first range R1 ranging from the average value Av ₁ to m · σ ₁ are determined as “normal”. And automatically selected as unit data D2. The data set containing outliers not enter the second range R2 ranging from the average value Av ₂ to k · sigma ₂ is excluded from the selection data D1. Therefore, even from the selection data D1 including the number of items of several hundred or more, it is possible to automatically obtain the unit data D2 by reducing the dependence by the selector only by setting the designated values m and k.

  In this embodiment, when the unit data D2 is replicated as the selection data D1, and the diagnosis target is diagnosed as normal, the diagnosis data is added to the selection data D1. As a result, the unit data D2 can be selected by automatically updating the selection data D1 at each abnormality diagnosis. As a result, for example, even when a parameter value regarded as normal data changes with time, or even when a data set is not sufficiently included in the selection data D1 when the unit space is initially generated, The unit data D2 can be optimized.

  In this embodiment, when a correction operation input is made via the input unit 30a of the data correction unit 30, the unit data D2 selected in S32 is corrected according to the operation input. As a result, the selected person can individually tune according to practical use (adding / deleting / correcting data in consideration of item-by-item distribution and human experience), and the diagnostic accuracy can be improved.

  The preferred embodiments of the present invention have been described above. However, the present invention is not limited to the above-described embodiments. The present invention is modified without departing from the scope described in the claims or applied to others. It may be.

For example, in the above-described embodiment, the first range R1 and the second range R2 are distributed around the average values Av ₁ and Av ₂ of the parameter values, but may be distributed around the mode value of the parameter values. Good. In the above embodiment, the step of excluding the data set including outliers from the selection data D1 (selection data excluding process) may be performed by a person.

  In the above embodiment, the unit data selection unit 21, the selection data exclusion unit 22, the diagnosis data acquisition unit 23, the abnormality diagnosis unit 24, and the selection data update unit 25 are configured by one CPU 20, but these At least one of them may be configured by different information processing apparatuses, or may be configured by different information processing apparatuses. Similarly, in the above embodiment, the selection data D1 and the unit data D2 are stored in the same storage device 10, but may be stored in different storage devices.

  In the above embodiment, the unit data selection unit 21, the selection data exclusion unit 22, the diagnosis data acquisition unit 23, the abnormality diagnosis unit 24, the selection data update unit 25, the data correction unit 30, the selection data D1, and the unit data D2 The input or output between each may be performed by wire or wirelessly.

21 unit data selection unit 21a first calculation unit 21b selection unit 22 selection data exclusion unit 22a second calculation unit 22b exclusion unit 23 diagnosis data acquisition unit 24 abnormality diagnosis unit 25 selection data update unit 100 abnormality diagnosis device d data set D1 Selection data D2 Unit data P Abnormality diagnosis program R1 First range R2 Second range

Claims (9)

Based on a data set having parameter values of a plurality of items related to a diagnosis object, using a diagnosis method on the assumption that the parameter value follows a normal distribution, an abnormality diagnosis method for diagnosing an abnormality of the diagnosis object,
From a selection data including a plurality of the data sets, comprising a data selection step of selecting unit data serving as a reference when performing diagnosis,
The data selection process includes
Calculating a first range based on a standard deviation of the parameter values for each of the plurality of items in the plurality of data sets included in the selection data;
Selecting, as the unit data, a data set in which the parameter value exists in the first range in all of the plurality of items among the plurality of data sets included in the selection data. Diagnosis method. A data excluding step of excluding a part of the selection data;
The data exclusion step includes
Calculating a kurtosis of the parameter value and a second range based on a standard deviation of the parameter value for each of the plurality of items in the plurality of data sets included in the selection data;
Removing the data set having the parameter value out of the second range from the plurality of data sets included in the selection data when the kurtosis is equal to or greater than the predetermined value. The abnormality diagnosis method according to 1. A data acquisition step of acquiring diagnostic data which is the data set for diagnosis from the diagnosis target;
Based on the unit data and the diagnostic data, a diagnostic step of diagnosing whether the diagnostic object is abnormal or normal;
A data update step of replicating the unit data as the selection data and adding the diagnostic data to the selection data when the diagnosis target is diagnosed as normal in the diagnosis step. 3. The abnormality diagnosis method according to 1 or 2. The diagnostic method is a Mahalanobis Taguchi method,
The abnormality diagnosis method according to claim 1, wherein the unit data constitutes a unit space that serves as a reference when calculating the Mahalanobis distance. Based on a data set having a plurality of parameter values related to a diagnosis target, using a diagnostic method based on the assumption that the parameter values follow a normal distribution, an abnormality diagnosis apparatus for diagnosing the abnormality of the diagnosis target,
A unit data selection unit that selects unit data serving as a reference when performing diagnosis from data for selection including a plurality of the data sets,
The unit data selection unit
A first calculator that calculates a first range based on a standard deviation of the parameter value for each of the plurality of items in the plurality of data sets included in the selection data;
A selection unit that selects, as the unit data, a data set in which the parameter value exists in the first range in all of the plurality of data sets included in the selection data; Abnormality diagnosis device. A selection data exclusion unit for excluding a part of the selection data;
The selection data excluding unit is
A second calculator that calculates a kurtosis of the parameter value and a second range based on a standard deviation of the parameter value for each of the plurality of data sets included in the selection data;
An exclusion unit that excludes a data set having the parameter value out of the second range from the plurality of data sets included in the selection data when the kurtosis is equal to or greater than the predetermined value. Item 6. The abnormality diagnosis device according to Item 5. A diagnostic data acquisition unit for acquiring diagnostic data which is the data set for diagnosis from the diagnostic target;
Based on the unit data and the diagnostic data, an abnormality diagnosis unit that diagnoses whether the diagnosis target is abnormal or normal;
A selection data update unit that replicates the unit data as the selection data and adds the diagnosis data to the selection data when the abnormality diagnosis unit diagnoses the diagnosis target as normal. The abnormality diagnosis device according to claim 5 or 6. The diagnostic method is a Mahalanobis Taguchi method,
The abnormality diagnosis apparatus according to claim 5, wherein the unit data constitutes a unit space that serves as a reference when calculating the Mahalanobis distance.   An abnormality diagnosis program for causing a computer to execute the abnormality diagnosis method according to claim 1.

Images