JP7233258B2 - Abnormal diagnosis device - Google Patents

Description

The present invention relates to an abnormality diagnosis device for plants or equipment.

Power plants and associated equipment are equipped with many sensors such as thermometers, pressure gauges and flow meters for monitoring and control purposes. In recent years, sensors such as acceleration sensors have also been installed in infrastructure facilities such as bridges and roads to monitor the status of facilities.

The feature of such sensor data is that there are multiple related sensor data, and it is multidimensional time-series data measured in a certain time period. Various methods have been proposed for condition monitoring methods utilizing such multidimensional time-series data.

For example, in Patent Document 1, as a technique using the adaptive resonance theory, "A plurality of measurement data from various sensors installed in the plant to be diagnosed belongs to the category discriminated by the normal data by the adaptive resonance theory. A plant abnormality diagnosis device characterized by comprising an abnormality degree calculation unit that calculates the abnormality degree of the entire plant to be diagnosed based on the difference in the spatial distance between the data and the plurality of measurement data. .

Here, adaptive resonance theory (ART: Adaptive Resonance Theory, hereinafter referred to as ART) is a technique for classifying multidimensional data into a plurality of categories according to similarity. . By using ART, anomalies can be detected by clustering multidimensional data and dividing it into categories.

WO2018/052568

G.A.Carpenter and S.Grossberg, "ART2: Self-OrganizationNo.f stable category recognition codes for analog input patterns", Applied Optics, Vol26, No.23, 1987

In Patent Literature 1, as shown in FIG. 1, an abnormality in equipment is detected based on operating data (diagnostic data) that deviates from a category (normal category) into which operating data (learning data) during normal operation is classified. In order to determine the degree of anomaly and which signal has the greatest deviation, the degree of anomaly and the degree of contribution to anomaly are calculated based on the difference between the normal data and the current measurement data.

Here, the degree of anomaly and the degree of contribution to anomaly in Patent Literature 1 will be described with reference to the drawings.

As shown in FIG. 2, when an abnormality is detected, the distance from the current data k to the center of gravity of the closest normal category (category 3 in the example of FIG. 2) is defined as the degree of abnormality. The degree of anomaly indicates the extent of data anomaly.

Further, as shown in FIG. 3, the degree of abnormality contribution is defined by decomposing the degree-of-abnormality vector into the components of each measurement value (x1 and x2 in the example of FIG. 3). The anomaly contribution indicates which signals deviate more from normal.

However, when using the degree of anomaly and the degree of contribution to anomaly described in Patent Document 1, there were cases where the degree of anomaly contribution changed significantly even though the measured data at the time of anomaly did not change significantly. An example thereof will be described with reference to FIG.

FIG. 4 shows a state in which new data m is measured in the vicinity in addition to the state in which data k shown in FIGS. 2 and 3 is abnormal. The nearest normal category for data k is category 3, while the nearest normal category for data m is category 1. As a result, the value of the degree of anomaly of both data does not change significantly, but the degree of contribution to anomaly changes greatly. That is, in data k, the degree of contribution to abnormality of x2 is large, but in data m, the degree of contribution to abnormality of data x1 is large.

As described above, in the method described in Patent Document 1, although the measurement data itself does not change significantly, the degree of contribution to abnormality may change significantly, and the cause of abnormality cannot be estimated based on the degree of contribution to abnormality. rice field.

In order to solve the above problems, an abnormality diagnosis apparatus according to the present invention includes a category calculation unit that classifies driving data output from a plurality of sensors provided on a diagnosis target into a plurality of categories; A data classification unit having a reference point calculation unit that calculates a reference point and an abnormality degree calculation unit that compares the reference point with current operation data and calculates the degree of abnormality of the current operation data, The reference point is obtained as a weighted average of centroid vectors of the normal category, and is set such that the coefficient of the weighted average decreases as the distance from the current data to the reference point increases.

Since the present invention has the above configuration, it is possible to improve the accuracy of estimating the cause of abnormality without a large change in the degree of contribution to abnormality in nearby measurement data.

A diagram showing how to classify conventional data. The figure which shows the example which calculated the degree of abnormality by the conventional method. The figure which shows the example which calculated the abnormality contribution by the conventional method. The figure which shows the subject at the time of calculating an abnormality contribution by the conventional method. The figure which shows the structure of the Example of this invention. The figure which shows the example of the driving data stored in the driving data database. The figure which shows the example of a structure of the data classification part of the Example of this invention. The figure which shows the calculation method of the degree of abnormality of this invention. The figure which shows an example of the trend graph of the classified category. The figure which shows an example of the trend graph of an abnormality degree and an abnormality contribution. The figure which shows test data. The figure which shows the example which calculated the degree of abnormality by the conventional method using test data. The figure which shows the example which calculated the abnormality contribution by the conventional method using test data. The figure which shows the example which calculated the degree of abnormality by the method of this invention using test data. The figure which shows the example which calculated the abnormality contribution with the method of this invention using test data.

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments of the present invention will be described with reference to the drawings. It should be noted that the following is merely an example of implementation, and the invention itself is not intended to be limited to the following specific contents.

FIG. 5 shows a configuration according to one embodiment of the invention. The present embodiment is an example in which the diagnosis apparatus of the present invention is used for plant abnormality diagnosis, and includes a plant 10 , an operation data database 20 , a data classification section 30 , a classification result database 40 and a display/operation section 50 . An outline of each component is described below. It should be noted that the target of abnormality diagnosis is not limited to plants, and can be applied to other devices.

The plant 10 is equipped with sensors and utilized for plant control and monitoring. Examples of sensors include, but are not limited to, flowmeters, thermometers, and pressure gauges.

The operation data database 20 stores the plant operation data measured by the sensors of the plant 10 as time-series data for each minute, for example. Of the stored operating data, the operating data designated as normal data by the display/operation unit 50 is cut out as learning data and sent to the data classification unit 30 . In addition, data measured in real time at the plant 10 is also temporarily stored in the operation data database 20, and then sent as diagnostic data to the data classification unit 30 at regular intervals.

The data classification unit 30 classifies multidimensional driving data into multiple categories using an ART-based clustering algorithm (hereinafter referred to as improved ART).

Here, ART is explained. ART classifies input data into multiple categories (clusters) and assigns a classified category number to each input data. A category represents a group of data having similarity, and input data to which the same category number is assigned has a high degree of similarity.

In the learning phase, the operating data (learning data) of the equipment in a normal state is input to ART. Since ART classifies the driving data (learning data) into multiple categories according to the similarity of the data, it defines the category (normal category) generated when the driving data is normal.

In the diagnosis phase, the driving data (diagnosis data) to be diagnosed is input to the ART that has learned normal data. As a result, data with a high degree of similarity to learning data are classified into the same category as the learning phase. However, when the tendency of the data changes, such as when some kind of abnormality occurs in the equipment, the data is classified into a category (new category) different from the learning data.

The data classification device 30 using improved ART according to the present embodiment outputs the category number into which the input multidimensional data is classified, the degree of abnormality, and the degree of contribution to abnormality of each measurement data as a result of data classification.

The classification result database 40 manages the category number, the degree of abnormality, and the degree of contribution to abnormality output from the data classification section 30 . In addition, weighting coefficients that serve as criteria for selection of each category and center-of-gravity data of data classified into each category are stored.

The display/operation unit 50 sets conditions for learning data and diagnosis data. Based on this condition, it is determined whether the operating data measured by the sensor of the plant 10 is learning data or diagnostic data. The display/operation unit 50 also displays a trend graph of the category number, the degree of abnormality, and the degree of contribution to abnormality. The display/operation unit 50 may be constructed by separating the display unit and the operation unit.

Next, the abnormality diagnosis method according to this embodiment will be described in detail with reference to FIGS. 6 to 10. FIG.

The plant 10 is composed of equipment and pipes, valves, and the like that connect the equipment. Equipment and piping are equipped with sensors such as flowmeters, thermometers, and pressure gauges for monitoring and controlling plant conditions. These sensors are tagged, for example, "TIC001" if they are thermometers. This tag becomes the ID of each sensor, and each sensor is identified by the ID.

The operation data database 20 records the data measured by the sensors of the plant 10 as time-series data. An example of operating data is shown in FIG. As shown in FIG. 6, as the operation data for each hour measured by the sensor installed in the plant 10, the time when the operation data was measured in the "Time" column, and "FIC001" and "PIC001" in the other columns. The value measured by the sensor specified by the ID such as is recorded. Although this time interval can be arbitrarily specified, it is set to 1 minute in this embodiment.

The data classification unit 30 classifies driving data using improved ART. A detailed configuration of the data classification unit 30 is shown in FIG. The data classification unit 30 includes a category calculation unit 31 , a reference value calculation unit 32 , an anomaly degree calculation unit 33 and an anomaly contribution degree calculation unit 34 .

The category calculation unit 31 classifies the driving data into categories according to the degree of similarity of the driving data. The category into which the driving data (learning data) is classified in the learning phase is the normal category. A detailed algorithm for classifying into categories is described in Non-Patent Document 1 and Patent Document 1, so the description is omitted here.

The reference point calculator 32 calculates a reference point for calculating the degree of abnormality. The degree of anomaly is calculated based on deviation information between the reference point and current driving data (diagnostic data).

A method of calculating the degree of abnormality according to the present invention will be described with reference to FIG. In this embodiment, the weighted average g _a of the centroids of the normal category is calculated by equation (1) as the reference point.

Here, g _i is the centroid vector (coordinates) of normal category j, and d _i ^(k) is the distance from g _i to the current measurement data k. f(x) is a monotonically decreasing function of x, and in this embodiment, the sigmoid function of equation (2) is used as an example.

According to equation (2), f(d _i ^(k) ) decreases as the distance d _i ^(k) increases. That is, the coefficient of the weighted average g _a is large for normal categories close to the measurement data m, and small for far categories. In particular, since the sigmoid function used in equation (2) is a function whose value asymptotically approaches zero as x increases, the coefficient of the category with a large distance d _i ^(k) approaches zero, affecting the calculation of the reference point. do not give

That is, the reference point obtained by Equation (1) is the average value of the centroids of the categories closest to the measurement data among the centroid coordinates of each category.

The degree-of-abnormality calculation unit 33 obtains the degree of abnormality of the current measurement data k from the coordinates of the reference point obtained by the reference point calculation unit 32 and the coordinates of the current measurement data k. The degree of anomaly is defined by the distance between the two, and the magnitude of the degree of anomaly vector d _a ^(k) in Equation (3) is the degree of anomaly.

The abnormality contribution calculator 34 calculates the abnormality contribution _{R i of} the parameter i by the following equation (4), with each component of the abnormality vector d _a ^(k) calculated by the abnormality calculator 33 as xi.

The classification result database 40 stores the category number, the degree of abnormality, and the degree of contribution to abnormality of each data obtained by the data classification unit 30 .

The display/operation unit 50 displays the data stored in the classification result database 40 and the operating data stored in the operating data database 20 . FIG. 9 shows a trend graph of classified categories. It can be seen that the category changes with time, and the state of the plant changes. In the example of FIG. 9, categories 1 to 3 are normal categories and category 4 is a new category.

FIG. 10 shows a display example of the degree of anomaly and the degree of contribution to anomaly at the same time as in FIG. The black line indicates the degree of anomaly. Since the degree of contribution to abnormality is obtained by Equation (4), the sum of the degrees of contribution to abnormality of each parameter is the degree of abnormality. The example shown in this figure is an example with two parameters, and the sum of the anomaly contribution _R1 of the parameter _x1 and the anomaly contribution _R2 of the parameter _x2 is the degree of anomaly. Comparing FIG. 9 and FIG. 10, it can be seen that the degree of abnormality increases from the time when the category becomes category 4, which is a new category.

Next, the results of comparing the conventional algorithm and the algorithm shown in this embodiment using test data will be described with reference to FIGS. 11 to 15. FIG.

FIG. 11 shows two-dimensional test data. There were 50 learning data points, and as a result of classification by ART, they were classified into 9 categories. The point illustrated with ♦ indicates the center of gravity of each category. There are 31 diagnostic data, and the data were entered in order from left to right at the time of diagnosis.

FIG. 12 and FIG. 13 show the results of obtaining the degree of anomaly and the degree of contribution to anomaly using a conventional algorithm. The degree of anomaly shown in FIG. 12 does not change smoothly at 7 minutes and 21 minutes, but there is no discontinuity, and there is no major problem in calculating the degree of anomaly even with the conventional algorithm. I understand. On the other hand, in the degree of contribution of abnormality shown in FIG. 13, the degree of contribution of abnormality changes discontinuously at the centroid of the closest normal category at the time of 7 minutes. If the abnormal contribution changes in such a discontinuous manner due to changes in the closest normal category, using this data to extrapolate the cause of the abnormality may be misleading.

Next, FIG. 14 and FIG. 15 show the results of obtaining the degree of anomaly and the degree of contribution to anomaly using the algorithm of the present invention. Compared to the case of using the conventional algorithm, the tendency of the degree of anomaly does not change much, but it can be seen that there is no discontinuous change in the graph of the degree of anomaly contribution.

As described above, by using the abnormality diagnosis apparatus of the present invention, the event that the degree of contribution to abnormality changes significantly despite the closeness of the measurement data can be resolved. That is, the accuracy is improved when estimating the cause of abnormality using the degree of contribution to abnormality.

In this embodiment, an example using ART as a data clustering technique is shown, but other clustering techniques may be used. Also, although the sigmoid function is used as f(x), another function such as 1/x may be used.

10: Plant 20: Operation data database 30: Data classification unit 31: Category calculation unit 32: Reference value calculation unit 33: Abnormality degree calculation unit 34: Abnormal contribution calculation unit 40: Classification result database 50: Display unit

Claims (9)

a category calculation unit that classifies driving data output from a plurality of sensors provided in a diagnosis target into a plurality of categories;
a reference point calculator that calculates reference points for the plurality of categories;
a degree-of-abnormality calculation unit that compares the reference point with the current operating data and calculates the degree of abnormality of the current operating data;
a data classifier having
The abnormality diagnosis apparatus, wherein the reference point is obtained as a weighted average of centroid vectors of normal categories, and is set such that the greater the distance from the current data to the reference point, the smaller the coefficient of the weighted average. 2. The abnormality diagnosis device according to claim 1, wherein the abnormality degree calculator calculates the degree of abnormality of the current operating data based on deviation information between the reference point and the current operating data. 3. The abnormality diagnosis device according to claim 1, wherein the degree-of-abnormality calculator calculates the degree of abnormality of the current operating data based on the distance between the reference point and the current operating data. 4. The abnormality diagnosis apparatus according to claim 3, wherein said operating data is multi-dimensional data composed of measurement data output from said plurality of sensors. 5. The abnormality diagnosis apparatus according to claim 4, further comprising an abnormality contribution degree calculation unit that calculates, as an abnormality contribution degree, the component of the degree of abnormality in the axial direction for each of the plurality of pieces of measurement data. an operating data database for outputting, from among the input operating data, operating data when the diagnosis target is normal as learning data to the data classification unit;
6. The abnormality diagnosis apparatus according to claim 5, wherein the category calculation unit classifies the category into which the learning data is classified as a normal category. 7. An abnormality diagnosis apparatus according to claim 6, wherein a weighted average coefficient is defined based on the distance from said current data to said reference point. 8. The abnormality diagnosis device according to claim 7, further comprising a display/operation unit that displays the classification results output from the data classification unit. 9. The abnormality diagnosis apparatus according to claim 8 , wherein the classification result includes the category number classified by the category calculation unit and the degree of abnormality.

Images