JP2016018435A - Parameter classification device - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve classification accuracy, facilitate classification, and achieving time reduction at a time of classifying clustering targets for abnormality diagnosis using the MT method.SOLUTION: A parameter classification device comprises: display processing means generating a time series graph representing values of a plurality of parameters in time series for parameter value acquisition, and a histogram of the parameter values, and displaying a display screen containing the generated time series graph and histogram on a display; and storage processing means associating an identifier of each parameter and a classification result with each other and storing the identifier of each parameter and the classification result in a storage device as classification data when receiving a result of classification as to whether each of the parameters for which the time series graph and the histogram are displayed on the display is a clustering target.SELECTED DRAWING: Figure 1

Description

  The present invention relates to a parameter classification apparatus used for classifying whether or not each parameter to be diagnosed is a clustering target when performing an abnormality diagnosis using the MT method.

  In recent years, the use of the MT method (Mahalanobis-Taguchi method), which is a type of MT system, for the diagnosis of abnormalities has become widespread. In the MT method, first, a unit space that is a reference for a normal state is generated from normal data that is a plurality of parameter values measured at the normal time from a device or system to be diagnosed, and a reference point of the unit space is determined. After that, the values of the same parameter are measured as the diagnosis data at the time of abnormality diagnosis, and abnormality is diagnosed using the distance between these values and the reference point (for example, refer to Non-Patent Document 1).

  When abnormality diagnosis is performed using the MT method, a clustering algorithm may be used. For example, the sensor value may change greatly before and after the switching due to the switching of the mode in the diagnosis target apparatus or system. In addition, the sensor value may change greatly before and after the change of the condition due to the change of the conditions such as temperature and season. In such a case, if the sensor value before the mode change and the sensor value after the mode change, and the sensor value before the change of the condition and the sensor value after the change are handled as the same, it is difficult to detect the abnormality. Therefore, since such parameters need to be handled as separate clusters before and after the change, they are selected as a clustering target, and an abnormality diagnosis is performed using a clustering algorithm.

  On the other hand, a continuous signal may actually be measured by a sensor or the like as a discrete signal. As a result, the measurement values of continuous signals become discrete because the sampling resolution of the sensor is low, etc., and the sensor values are divided into a small number of groups and may be easily misunderstood as a clustering target.

Tatebayashi, Teshima, Hasegawa, “Introductory MT System”, Nikka Tech Ren Publisher, 2008, 43-44

  As described above, the parameter may be measured as a continuous signal or a discrete signal, and it is difficult to determine whether or not it is a clustering target. For this reason, the accuracy of classification of each parameter into clusters depends on the experience of the operator. In addition, since clustering targets are classified by referring to values measured multiple times for each parameter, there is a current situation that requires a long time for classification when the number of parameters is large.

  In view of the above problems, the present invention improves the accuracy of classification and facilitates the processing and shortens the time required for classification when classifying objects for clustering for abnormality diagnosis using the MT method. Objective.

  In order to achieve the above object, a first invention is a parameter classification device used for classifying whether or not each parameter to be diagnosed is a clustering target when performing an abnormality diagnosis using the MT method. Display processing means for generating a time series graph representing parameter values for a plurality of times in a time series of parameter value acquisition and a histogram of the parameter values, and displaying a display screen including the generated time series graph and the histogram on a display; Upon receiving the result of the classification of whether or not the parameter whose time series graph and histogram are displayed on the display is the target of clustering, the identifier of the parameter and the classification result are associated and stored in the storage device as classification data. Storage processing means.

  According to the present invention, when classifying a clustering target for abnormality diagnosis using the MT method, the accuracy of classification can be improved, the processing can be facilitated, and the time required for classification can be shortened.

FIG. 1 is a block diagram illustrating a parameter classification device according to an embodiment. 2A is a scatter diagram of two parameters without a cluster structure, and FIG. 2B is a scatter diagram of two parameters with a cluster structure. FIG. 3 is a graph illustrating an example of the relationship between the actual signal of the parameter and the value measured by the sensor. FIG. 4 is an example of a screen displayed by the parameter classification device of FIG. 1 when the values of FIG. 3 are measured. FIG. 5 is a graph for explaining another example of the relationship between the actual signal of the parameter and the value measured by the sensor. FIG. 6 is an example of a screen displayed by the parameter classification device of FIG. 1 when the values of FIG. 5 are measured. FIG. 7 is a graph for explaining another example of the relationship between the actual signal of the parameter and the value measured by the sensor. FIG. 8 is an example of a screen displayed by the parameter classification device of FIG. 1 when the values of FIG. 7 are measured.

  Hereinafter, a parameter classification device according to an embodiment of the present invention will be described with reference to the drawings. The parameter classification apparatus according to the embodiment uses the MT method to perform abnormality diagnosis using each parameter value for gas turbines, vacuum furnaces, aircraft engines, and the like whose parameter values are measured by a large number of sensors. When performing, each parameter value is used to classify whether or not it is a clustering target. Note that the parameter classification device according to the embodiment may be incorporated in an abnormality diagnosis device that detects an abnormality of a diagnosis target. For parameters classified as clustering targets, a clustering algorithm is used for abnormality diagnosis, a unit space is formed for each cluster, and newly measured parameter values are associated with one of the clusters. Abnormal diagnosis is executed.

  As shown in FIG. 1, the parameter classification device 1 according to the embodiment includes a CPU 10, a storage device 20, an input device 30 used for operation input and the like, an output device 40 used for outputting processing results, and the like. Information processing apparatus.

  As shown in FIG. 1, the parameter classification device 1 is obtained by executing the classification program P stored in the storage device 20 so that the CPU 20 acquires the value (parameter value) of each item to be diagnosed. The display processing means 12 for displaying the screen used for classification on the output device 40, the receiving means 13 for receiving the classification result input via the input device 30, and the received classification result are stored in the storage device 20. Processing is executed as the storage processing means 14.

  In addition to the classification program P, the storage device 20 stores target data D1 that is each item value to be diagnosed, screen data D2 that is used to display a screen, and classification data D3 that is a classification result.

  The target data D1 includes a plurality of measured item values.

  FIG. 2 is an example of a scatter diagram regarding the values of certain two parameters. In FIG. 2, the horizontal axis is the value of parameter 1 (item value 1), and the vertical axis is the value of parameter 2 (item value 2). An example shown in FIG. 2A is a scatter diagram regarding item values that do not require clustering. When there is no need for clustering, as shown in FIG. 2A, the distribution of values forms a homogeneous group.

  On the other hand, the example shown in FIG. 2B is a scatter diagram regarding item values that need to be clustered. In the example shown in FIG. 2B, the distribution of all values does not form a homogeneous group but is divided into two groups. In such a case, when performing abnormality diagnosis by the MT method, it is necessary to classify the data of cluster 1 and cluster 2 using a clustering algorithm and diagnose abnormality separately.

  FIG. 3 is an example of time-series data of items whose item values are divided into a plurality of groups because the sampling resolution of the sensor is low. In FIG. 3, the horizontal axis represents the number of measurements (number numbered for each number of measurements), the vertical axis represents the item value (parameter value), and the measured item value is represented in time series. The example shown in FIG. 3 is an example in which the actual signal (solid line in FIG. 3) is continuous, whereas the measurement value by the sensor (broken line in FIG. 3) is discrete.

  Specifically, it is assumed that the average value of signals measured in the normal state is 6.5, the amplitude is 0.1, and the resolution of the sensor is 1. At this time, the sensor outputs a value of either 7 or 6 depending on whether the signal to be measured is greater than or less than 6.5. Thus, while the actual signal is continuous, the sensor measurements appear to form two clusters. In such a case, group 1 and group 2 are not actually separate clusters, but are erroneously determined as separate clusters, and this item is likely to be erroneously classified as a clustering target. If classified incorrectly in this way, a clustering algorithm is applied to items that are not subject to clustering, and the accuracy of abnormality diagnosis decreases. Therefore, it is preferable to accurately classify the clustering target.

  The acquisition unit 11 acquires item values (parameter values) for classifying whether or not clustering is necessary. For example, the acquisition unit 11 may acquire an item value from a sensor to be diagnosed or the like, or may acquire an item value from an abnormality diagnosis device. The acquisition unit 11 stores the acquired item values in the storage device 20 as the target data D1.

  The display processing means 12 reads the target data D1 and the screen data D2 from the storage device 20, generates a display screen used for classification as to whether or not it is a clustering target, and displays it on the output device 40. For example, the display processing means 12 generates a display screen W as shown in FIG.

  Specifically, the display processing unit 12 generates a time-series graph of target item values using the target data D1. Further, the display processing means 12 uses the target data D1 to generate a histogram that is a frequency distribution measured for the target item value. Thereafter, the display processing unit 12 combines the screen data D2 with the generated time series graph and histogram to form the display screen W.

  The display screen W is displayed for each target item. As shown in FIG. 4, a graph display area A1 including a time-series graph, a histogram display area A2 including a histogram of item values, and an operator can input comments. A comment entry area A3 is provided. The display screen W displays a first button B1 that is pressed when it is determined to be a clustering target, a second button B2 that is pressed when it is determined that it is not a clustering target, and a display screen related to the previous item. A third button B3 and a fourth button B4 for displaying a display screen relating to the next item are provided.

  FIG. 4 is an example of the display screen W related to the item with the item name “item ΔΔ”. The item “item ΔΔ” is the x-th item among the y items. The operator who refers to the display screen W easily determines whether or not it is a clustering target by referring to the time-series graph displayed in the graph display area A1 and the histogram displayed in the histogram display area A2. be able to. Depending on the result of such determination, the first button B1 or the second button B2 is pressed.

  For example, as shown in the histogram display area A2 in FIG. 4, when the histogram is a vertical bar with a width distributed at a constant interval, it is considered as a group depending on the sampling performance of the sensor and is not a clustering target. It is judged. Therefore, in the example shown in FIG. 4, the second button B2 is pressed. Further, the example shown in the histogram display area A2 in FIG. 4 is an example including only two item values, but the same applies to the case including two or more item values. If a histogram representing an item that depends on the sampling performance of the sensor includes two or more item values, the item values are separated at a constant interval.

  On the other hand, even if a group of item values is formed in the histogram, if the distribution of the frequency of each item value group is continuous, it is determined that it is a clustering target. When subject to clustering, for example, as shown in FIG. 5, the time-series data is not continuous as the actual signal (solid line in FIG. 5) and the sensor measurement value (broken line in FIG. 5) are not discrete. Is. Specifically, in the actual signal shown in FIG. 5, the average values of cluster 1 and cluster 2 are 36.5 and 6.5, respectively, and the amplitude is 6. The measurement values shown in FIG. 5 are assumed to be rounded to an integer when the sampling resolution of the sensor is 1. In the example shown in FIG. 5, since the sampling resolution is sufficient as compared with the example shown in FIG. 3, the actual signal and the measured value of the sensor are almost the same.

  When time-series data as shown in FIG. 5 is obtained, the display processing means 12 generates a display screen W by combining the time-series graph and histogram as shown in FIG. In the example shown in FIG. 6, a plurality of item value groups are formed in the histogram. However, since each group has a different frequency and is a clustering target, the first button B1 is pressed.

  If a group of item values is not formed in the histogram, it is determined that it is not a clustering target. When a group of item values is not formed in the histogram, time series data is generated as shown in FIG. 7, for example. Specifically, in the actual signal shown in FIG. 7, the average value of the measurement values at the normal time is 6.5. The measurement values shown in FIG. 7 are assumed to be rounded to an integer when the sampling resolution of the sensor is 1.

  When time-series data as shown in FIG. 7 is obtained, the display processing unit 12 generates a display screen W by combining the time-series graph and the histogram as shown in FIG. In the example shown in FIG. 8, since there is one item value group in the histogram and it is understood that it is not a clustering target, the second button B <b> 2 is pressed.

  In the comment entry area A3, the operator can freely enter matters regarding this item via the input device 30 based on the graph display area A1 and the histogram display area A2. For example, an abnormality appearing in a sensor that measures a target item value, a phenomenon to be noted, and the like can be entered.

  The receiving unit 13 receives a clustering selection signal when the first button B1 is pressed via the input device 30 after the display processing unit 12 displays the display screen W. When the receiving means 13 receives the clustering selection signal, the storage processing means 14 classifies this item as a clustering target, and updates the classification data D3 by associating the identifier of this item with the classification result that is the clustering target.

  The receiving unit 13 receives a non-selection signal when the second button B2 is pressed through the input device 30 after the display processing unit 12 displays the display screen W. When the receiving unit 13 receives the non-selection signal, the storage processing unit 14 classifies this item as non-clustering target, and updates the classification data D3 by associating the identifier of this item with the classification result that is non-clustering target. .

  Further, when the receiving means 13 receives the contents entered in the comment entry area A3, the storage processing means 14 adds the contents received by the receiving means 13 to the classification data D3. Thereby, the operator who performed the classification work can share the comment information entered in the comment entry area A3 with other operators.

  The receiving means 13 receives the previous item signal when the third button is pressed via the input device 30 after the display processing means 12 displays the display screen W, and receives the next item signal when the fourth button is pressed. To do. The receiving unit 13 outputs the received previous item signal or next item signal to the display processing unit 12.

  When the previous item signal or the next item signal is input from the receiving unit 13, the display processing unit 12 generates a display screen W of the corresponding item according to the signal and displays it on the output device 40. For example, when the previous item signal is received after the display screen W shown in FIG. 4 is displayed, the display processing means 12 displays the display screen W regarding the (x−1) th item. In addition, when the rear item signal is received after the display screen W shown in FIG. 4 is displayed, the display processing means 12 displays the display screen W regarding the (x + 1) th item.

As described above, in the parameter classification device according to the embodiment, whether or not each parameter (item) is a clustering target is classified using the display screen W. Therefore, the operator who classifies the parameters can be easily classified from the visual information displayed on the display screen W.
As mentioned above, although this invention was demonstrated in detail using embodiment, this invention is not limited to embodiment described in this specification. The scope of the present invention is determined by the description of the claims and the scope equivalent to the description of the claims.

DESCRIPTION OF SYMBOLS 1 Parameter classification | category apparatus 11 Acquisition means 12 Display processing means 13 Reception means 14 Storage processing means 20 Storage apparatus 30 Input apparatus 40 Output apparatus D1 Target data D2 Screen data D3 Classification data P Classification program

Claims (1)

A parameter classification device used for classifying whether or not each parameter to be diagnosed is a clustering target when performing an abnormality diagnosis using the MT method,
Display processing means for generating a time series graph representing a parameter value acquisition time series and a histogram of the parameter value, and displaying a display screen including the generated time series graph and the histogram on a display;
When a result of categorizing whether or not it is a clustering target for a parameter whose time series graph and histogram are displayed on the display is stored, the identifier of the parameter and the classification result are associated with each other and stored in the storage device as classification data Processing means;
A parameter classification device comprising:

Images