Abnormality detector

Abstract

To realize a configuration that can reduce calculation loads in an abnormality detector that detects an abnormality of a target device by use of teacher data.SOLUTION: An abnormality detector 1 has: a normal state partial data extraction part 12 that extracts a plurality of pieces of normal state partial data from normal state continuous data; a similarity determination part 13 that determines similar normal state partial data in the plurality of pieces of normal state partial data extracted by the normal state partial data extraction part 12; a teacher data generation part 14 that generates teacher data by omitting the similar normal state partial data determined to be similar in the plurality of pieces of normal state partial data by the similarity determination part 13; a measurement partial data extraction part 22 that extracts a plurality of pieces of measurement partial data from measurement continuous data; and an abnormality level calculation part 15 that calculates an abnormality level according to a degree of similarity between the measurement partial data and the normal state partial data included in the teacher data.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to an abnormality detection device that detects abnormalities in a device.

Background Art

[0002] An abnormality detection device that detects abnormalities in a target device is known. As such an abnormality detection device, for example, Patent Document 1 discloses an abnormality detection device that performs a determination on the occurrence of an abnormality using abnormality determination information stored in an abnormality determination information storage unit. The abnormality determination information is information in which a system parameter and an abnormality determination condition for determining an abnormality are associated.

[0003] Patent Document 1 discloses that the abnormality determination information may include information obtained using a machine learning algorithm such as the k-nearest neighbor method.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] By the way, when using the k-nearest neighbor method to detect an abnormality in a target device, in order to detect the abnormality, not only measurement continuous data obtained by measuring a physical quantity of the target device but also teacher data is required. This teacher data is acquired based on normal-time continuous data of the physical quantity of the target device. For example, the teacher data is obtained by extracting a plurality of normal-time partial data while shifting the time within a predetermined range from the normal-time continuous data. Therefore, the number of the normal-time partial data used for the teacher data is quite large.

[0006] When using the k-nearest neighbor method to detect an abnormality in a target device, it is necessary to calculate the degree of abnormality of a plurality of measured partial data extracted from the measured continuous data for the normal-time partial data. Therefore, if the number of the normal-time partial data and the measured partial data is large, the amount of calculation increases, and the calculation load on the abnormality detection device increases. In particular, when the measurement data and the teacher data include data related to a plurality of states (for example, startup, operation, stop, etc.) of the target device, the amount of data further increases, and the calculation load on the abnormality detection device further increases.

[0007] An object of the present invention is to realize a configuration capable of reducing the calculation load in an abnormality detection device that detects an abnormality of a target device using teacher data.

Means for Solving the Problem

[0008] An abnormality detection device according to an embodiment of the present invention is an abnormality detection device that detects an abnormality of a target device using teacher data obtained based on normal-time continuous data related to a physical quantity of the target device and measurement continuous data obtained by measuring the physical quantity of the target device. This abnormality detection device includes a normal-time partial data extraction unit that extracts a plurality of normal-time partial data from the normal-time continuous data, a similarity determination unit that determines similar normal-time partial data among the plurality of normal-time partial data extracted by the normal-time partial data extraction unit, a teacher data generation unit that generates the teacher data by omitting the similar normal-time partial data determined to be similar by the similarity determination unit among the plurality of normal-time partial data, a measurement partial data extraction unit that extracts a plurality of measurement partial data from the measurement continuous data, and an abnormality degree calculation unit that calculates an abnormality degree according to the similarity between the normal-time partial data included in the teacher data and the measurement partial data (first configuration).

[0009] As a result, among a plurality of normal-time partial data extracted from the normal-time continuous data regarding the physical quantity of the target device, similar normal-time partial data can be omitted to generate the teacher data. Thus, the normal-time partial data in the teacher data can be shortened.

[0010] Therefore, in an anomaly detection device that detects an anomaly of a target device using teacher data, a configuration capable of reducing the calculation load can be realized.

[0011] In the first configuration, when the anomaly degree calculated by the anomaly degree calculation unit is equal to or greater than a predetermined value, the anomaly detection device further includes an anomaly detection unit that detects an anomaly of the target device (second configuration). Thereby, based on the anomaly degree calculated by the anomaly degree calculation unit, an anomaly of the target device can be detected. Moreover, by determining the anomaly of the target device according to whether the anomaly degree is equal to or greater than the predetermined value, the anomaly of the target device can be easily detected.

[0012] In the second configuration, the similarity determination unit determines that the two normal-time partial data are similar when the similarity between the two normal-time partial data is equal to or less than a threshold value. The threshold value is smaller than the predetermined value (third configuration).

[0013] Thereby, the similarity between two normal-time partial data can be determined, and according to the setting of the threshold value, the amount of data to be omitted in the plurality of normal-time partial data can be adjusted. Moreover, since the threshold value is smaller than the predetermined value used for determining whether there is an anomaly, measurement partial data that is clearly different from the normal-time partial data in the plurality of normal-time partial data can be more reliably detected as an anomaly.

[0014] In the first configuration, when the teacher data generation unit can ensure the continuity between the normal-time partial data located before the similar normal-time partial data and the normal-time partial data located after the similar normal-time partial data in the state where the plurality of normal-time partial data are arranged in the generation order, the teacher data generation unit omits the similar normal-time partial data to generate the teacher data (fourth configuration).

[0015] Thereby, even when the normal-time partial data determined to be similar by the similarity determination unit is omitted in the plurality of normal-time partial data, the continuity in the plurality of normal-time partial data after omission can be ensured. Thereby, the teacher data can be restored to continuous data. That is, with the above-described configuration, reversible processing of the teacher data can be performed. Thereby, after generating the teacher data, the conditions for extracting the plurality of normal-time partial data from the normal-time continuous data can be changed to adjust the data amount and accuracy of the teacher data, and the degree of freedom in generating the teacher data can be improved. Moreover, the plurality of normal-time partial data after omission can be visualized by a graph or the like.

[0016] In the fourth configuration, when the rear data of the normal-time partial data located before the similar normal-time partial data matches the front data of the normal-time partial data located after the similar normal-time partial data, the teacher data generation unit omits at least the similar normal-time partial data to generate the teacher data (fifth configuration).

[0017] Thereby, in the plurality of normal-time partial data, the partial data determined to be similar by the similarity determination unit can be omitted while ensuring continuity.

Advantages of the Invention

[0018] An abnormality detection device according to an embodiment of the present invention includes a normal-time partial data extraction unit that extracts a plurality of normal-time partial data from normal-time continuous data, a similarity determination unit that determines similar normal-time partial data among the plurality of normal-time partial data extracted by the normal-time partial data extraction unit, a teacher data generation unit that generates the teacher data by omitting the similar normal-time partial data determined to be similar by the similarity determination unit among the plurality of partial data, a measurement partial data extraction unit that extracts a plurality of measurement partial data from measurement continuous data, and an abnormality degree calculation unit that calculates an abnormality degree according to the similarity between the normal-time partial data included in the teacher data and the measurement partial data.

[0019] As a result, since similar normal-time partial data can be omitted to generate teacher data, the data volume of the teacher data can be reduced. Therefore, in an abnormality detection device that calculates an abnormality degree according to the similarity between the normal-time partial data included in the teacher data and a plurality of measurement partial data extracted from measurement continuous data, the calculation load can be reduced.

Brief Description of Drawings

[0020]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Embodiment for Carrying Out the Invention

[0021] Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings. The same or corresponding parts in the drawings are denoted by the same reference numerals and their descriptions will not be repeated.

[0022] [Embodiment 1] FIG. 1 is a diagram showing a schematic configuration of an abnormality detection device 1 according to Embodiment 1 of the present invention in functional blocks. This abnormality detection device 1 is a device that detects an abnormality of a target device P such as a vibration conveying device, a vibration generating device, or a press machine based on teacher data TD and measurement continuous data MD which is continuous data measured by a sensor or the like. The abnormality detection device 1 may be realized by an arithmetic device such as a computer or a server, or may be realized by a device in which a dedicated circuit is formed.

[0023] The measurement continuous data MD may be data of a single type of physical quantity and at the same measurement position, or may be data of a plurality of types of physical quantities or data with different measurement positions.

[0024] The abnormality means a state different from normal. The continuous data is measurable data and means a plurality of continuously arranged data. The continuous data may be, for example, data arranged in time series, or data arranged in a predetermined order according to a measurement position, a physical quantity, or the like.

[0025] The abnormality detection device 1 detects an abnormality by using normal-time continuous data ND, which is past normal continuous data regarding the physical quantity of the target device P. Specifically, the abnormality detection device 1 extracts a plurality of normal-time partial data Nd from the normal-time continuous data ND to obtain teacher data TD, and obtains the degree of abnormality by calculating the distance between the teacher data TD and the data to be inspected (measurement partial data Md extracted from the measurement continuous data MD regarding the physical quantity of the target device P). Specifically, the abnormality detection device 1 calculates the distance between one measurement partial data Md and all the normal-time partial data Nd, and obtains the minimum distance among the calculation results as the degree of abnormality. FIG. 2 is a diagram schematically showing the calculation of the degree of abnormality by the abnormality detection device 1. The abnormality detection device 1 detects the presence or absence of an abnormality in the target device P based on the calculated degree of abnormality.

[0026] As shown in FIG. 1, the abnormality detection device 1 includes a normal-time continuous data acquisition unit 11, a normal-time partial data extraction unit 12, a similarity determination unit 13, a teacher data generation unit 14, an abnormality degree calculation unit 15, an abnormality detection unit 16, a measurement continuous data acquisition unit 21, and a measurement partial data extraction unit 22.

[0027] The normal-time continuous data acquisition unit 11 acquires past normal-time continuous data ND regarding the physical quantity of the target device P. The physical quantity of the target device P means a physical quantity related to the target device P. That is, the physical quantity of the target device P may be, for example, a physical quantity indicating the operating state of the target device P, or a physical quantity in the output of the target device P. The physical quantity of the target device P is measured in time series by, for example, a sensor or the like.

[0028] The normal-time continuous data acquisition unit 11 acquires the normal-time continuous data ND regarding the physical quantity of the target device P measured by, for example, a sensor or the like from the sensor or the like. Note that the normal-time continuous data ND of the physical quantity of the target device P may be stored in a storage device such as a memory. In this case, the normal-time continuous data acquisition unit 11 acquires the past normal-time continuous data ND regarding the physical quantity of the target device P from the storage device.

[0029] During normal operation, the normal-time partial data extraction unit 12 extracts a plurality of normal-time partial data Nd within a predetermined data range from the normal-time continuous data ND acquired by the normal-time continuous data acquisition unit 11 while shifting them in the data arrangement direction. Specifically, as shown in FIG. 2, the normal-time partial data extraction unit 12 uses a sliding window S to extract the normal-time partial data Nd from the normal-time continuous data ND so as to fall within the predetermined data range. At this time, the normal-time partial data extraction unit 12 extracts a plurality of normal-time partial data Nd from the normal-time continuous data ND while shifting the sliding window S in the data arrangement direction. For example, when the normal-time continuous data ND is data arranged in time series, the normal-time partial data extraction unit 12 extracts a plurality of normal-time partial data Nd from the normal-time continuous data ND while shifting the sliding window S by a predetermined time.

[0030] Thereby, a plurality of normal-time partial data Nd can be extracted from the normal-time continuous data ND.

[0031] The similarity determination unit 13 shown in FIG. 1 determines whether the normal-time partial data Nd are similar to each other among the plurality of normal-time partial data Nd. The similarity determination unit 13 determines that one normal-time partial data Nd is similar to the other normal-time partial data Nd, for example, when the similarity between two normal-time partial data Nd is equal to or less than a threshold value. The threshold value is set to a value of the similarity of the normal-time partial data Nd such that it is considered that there is no problem in generating teacher data by omitting the similar normal-time partial data Nd among the plurality of normal-time partial data Nd to be compared.

[0032] The similarity determination unit 13 obtains the similarity using, for example, the k-nearest neighbor method, DTW (Dynamic Time Warping), the mean squared error, etc. The similarity is a value representing the degree to which two normal-time partial data Nd are similar. When the similarity is zero, the two normal-time partial data Nd are exactly the same. The greater the similarity, the more distant the two normal-time partial data Nd are from each other. Since the normal-time partial data Nd and the measurement partial data Md are real numbers, not integers, the similarity determination unit 13 needs to determine whether two normal-time partial data Nd are similar using the similarity as described above.

[0033] Note that the similarity determination unit 13 may obtain the similarity using fitting statistics such as the mean squared error, the mean absolute error, or the coefficient of determination, or may obtain it using a method for calculating the similarity between vectors such as the Euclidean distance, Pearson's correlation coefficient, or deviation pattern similarity.

[0034] The teacher data generation unit 14 generates teacher data TD by omitting the normal-time partial data (similar normal-time partial data Ndd) determined to be similar by the similarity determination unit 13 from among a plurality of normal-time partial data Nd. Thereby, the data amount of the teacher data TD generated by the teacher data generation unit 14 can be made smaller than the data amount of the plurality of normal-time partial data Nd extracted from the normal-time continuous data ND.

[0035] FIG. 3 is a diagram showing an example of obtaining teacher data TD by extracting a plurality of normal-time partial data Nd from the normal-time continuous data ND using a sliding window S and omitting similar normal-time partial data Ndd that are similar. As shown in FIG. 3, among the plurality of normal-time partial data Nd extracted from the normal-time continuous data ND using the sliding window S, the overlapping similar normal-time partial data Ndd (data described in italic characters in FIG. 3) are omitted to generate the teacher data TD. Thereby, the number of normal-time partial data Nd included in the teacher data TD can be reduced, and the data amount of the teacher data TD can be reduced.

[0036] The continuous measurement data acquisition unit 21 shown in FIG. 1 acquires continuous measurement data MD obtained by measuring the physical quantity of the target device P. The continuous measurement data acquisition unit 21 may acquire, as the continuous measurement data MD, data regarding the physical quantity of the target device P measured by, for example, a sensor or the like. The continuous measurement data acquisition unit 21 may acquire, as the continuous measurement data MD, data regarding the physical quantity of the target device P stored in a storage device such as a memory, for example.

[0037] The measurement partial data extraction unit 22 extracts a plurality of measurement partial data Md within a predetermined data range from the continuous measurement data MD acquired by the continuous measurement data acquisition unit 21 while shifting in the data arrangement direction. Specifically, as shown in FIG. 2, the measurement partial data extraction unit 22 uses a sliding window S to extract the measurement partial data Md from the continuous measurement data MD so as to fall within the predetermined data range. At this time, the measurement partial data extraction unit 22 extracts a plurality of measurement partial data Md from the continuous measurement data MD while shifting the sliding window S in the data arrangement direction. For example, when the continuous measurement data MD is data arranged in time series, the measurement partial data extraction unit 22 extracts a plurality of measurement partial data Md from the continuous measurement data MD while shifting the sliding window S by a predetermined time.

[0038] The abnormality degree calculation unit 15 calculates the abnormality degree using the teacher data TD generated by the teacher data generation unit 14 and the measurement partial data Md extracted from the continuous measurement data MD of the target device P. Specifically, as shown in FIG. 2, the abnormality degree calculation unit 15 calculates the distance between the normal-time partial data Nd included in the teacher data TD and the measurement partial data Md using the k-nearest neighbor method. Specifically, the abnormality degree calculation unit 15 calculates the distance between one measurement partial data Md and all the normal-time partial data Nd using the k-nearest neighbor method, and obtains the minimum distance among the calculation results as the abnormality degree.

[0039] Note that the abnormality degree calculation unit 15 may calculate the similarity between one measurement partial data Md and all the normal-time partial data Nd by the same method as the similarity determination unit 13, and obtain the similarity in the case of the most similar among the calculation results as the abnormality degree.

[0040] The abnormality detection unit 16 shown in FIG. 1 detects an abnormality of the target device P by using the degree of abnormality calculated by the degree-of-abnormality calculation unit 15. Specifically, when the degree of abnormality is equal to or greater than a predetermined value, the abnormality detection unit 16 detects that the target device P is abnormal. The result of the abnormality detection by the abnormality detection unit 16 is output as an abnormality detection signal. Note that the output abnormality detection signal may be used for display on a display device, may be used for notification by sound, light, or the like, or may be used for other information processing.

[0041] Note that the predetermined value is greater than the threshold value for determining the similarity of the normal-time partial data Nd during normal operation. That is, the threshold value is smaller than the predetermined value. Also, the predetermined value may be determined based on the average of the degrees of abnormality of the target device P during normal operation.

[0042] The abnormality detection device 1 according to the present embodiment detects an abnormality of the target device P by using teacher data TD obtained based on normal-time continuous data ND regarding a physical quantity of the target device P and measurement continuous data MD obtained by measuring the physical quantity of the target device P. This abnormality detection device 1 includes a normal-time partial data extraction unit 12 that extracts a plurality of normal-time partial data Nd from the normal-time continuous data ND, a similarity determination unit 13 that determines similar normal-time partial data Nd among the plurality of normal-time partial data Nd extracted by the normal-time partial data extraction unit 12, a teacher data generation unit 14 that generates the teacher data TD by omitting the similar normal-time partial data determined to be similar by the similarity determination unit 13 among the plurality of normal-time partial data Nd, a measurement partial data extraction unit 22 that extracts a plurality of measurement partial data Md from the measurement continuous data MD, and an abnormality degree calculation unit 15 that obtains the distance between the normal-time partial data Nd included in the teacher data TD and the measurement partial data Md by using the k-nearest neighbor method and calculates the degree of abnormality according to the distance.

[0043] As a result, among a plurality of normal-time partial data Nd extracted from the normal-time continuous data ND regarding the physical quantity of the target device P, similar normal-time partial data Ndd can be omitted to generate the teacher data TD. Therefore, the data volume of the teacher data TD can be reduced.

[0044] Further, when the abnormality degree calculated by the abnormality degree calculation unit 15 is equal to or greater than a predetermined value, it further includes an abnormality detection unit 16 that detects an abnormality of the target device P. Thereby, based on the abnormality degree calculated by the abnormality degree calculation unit 15, an abnormality of the target device P can be detected. Moreover, by determining the abnormality of the target device P according to whether the abnormality degree is equal to or greater than the predetermined value, the abnormality of the target device P can be easily detected.

[0045] In addition, the similarity determination unit 13 determines that two normal-time partial data Nd are similar when the similarity degree between the two normal-time partial data Nd is equal to or less than a threshold value. The threshold value is smaller than the predetermined value.

[0046] As a result, the similarity between two normal-time partial data Nd can be determined, and according to the setting of the threshold value, the data volume to be omitted among the plurality of normal-time partial data Nd can be adjusted. Moreover, since the threshold value is smaller than the predetermined value used for determining whether it is abnormal, clearly different measurement partial data Md can be more reliably detected as abnormal for the plurality of normal-time partial data Nd.

[0047] [Embodiment 2] FIG. 4 is a diagram showing a schematic configuration of the abnormality detection device 100 according to Embodiment 2. The abnormality detection device 100 is different from the abnormality detection device 1 of Embodiment 1 in that it has a continuity confirmation unit 115. Hereinafter, the same parts as those in Embodiment 1 are denoted by the same reference numerals and the description thereof is omitted, and only the parts different from those in Embodiment 1 will be described.

[0048] As shown in FIG. 4, the abnormality detection device 100 includes a normal continuous data acquisition unit 11, a normal partial data extraction unit 12, a similarity determination unit 13, a continuity confirmation unit 115, a teacher data generation unit 114, an abnormality degree calculation unit 15, an abnormality detection unit 16, a measurement continuous data acquisition unit 21, and a measurement partial data extraction unit 22.

[0049] The continuity confirmation unit 115 confirms the continuity of the normal partial data Nd located before and after the normal partial data (similar normal partial data Ndd) determined to be similar by the similarity determination unit 13 in a state where a plurality of normal partial data Nd are arranged in the order of data generation. Specifically, the continuity confirmation unit 115 determines whether the latter half part (reference F in FIG. 5) of the normal partial data Nd located before the similar normal partial data Ndd determined to be similar by the similarity determination unit 13 matches the former half part (reference R in FIG. 5) of the normal partial data Nd located after the similar normal partial data Ndd. Specifically, the continuity confirmation unit 115, similar to the similarity determination unit 13 in Embodiment 1, calculates the similarity between the data of the latter half part of the normal partial data Nd located before the similar normal partial data Ndd and the data of the former half part of the normal partial data Nd located after the similar normal partial data Ndd, and determines that they match when the calculated similarity is equal to or less than a threshold value. Since the calculation and determination of the similarity are the same as those in Embodiment 1, detailed description thereof is omitted.

[0050] The latter half part and the former half part are part of the partial data within a predetermined range determined by the sliding window S. Specifically, the latter half part and the former half part include data that is one less than the number of data determined by the window width (length) of the sliding window S in the normal partial data Nd. The number of data included in the latter half part and the former half part is the same.

[0051] When the continuity confirmation unit 115 determines that the normal-time partial data Nd located before and after the similar normal-time partial data Ndd has continuity, the teacher data generation signal generation unit generates a teacher data generation signal for causing the teacher data generation unit 114 to generate the teacher data TD by omitting the similar normal-time partial data Ndd. The teacher data generation signal is input to the teacher data generation unit 114.

[0052] The teacher data generation unit 114 generates the teacher data TD based on the teacher data generation signal output from the continuity confirmation unit 115. Specifically, when the teacher data generation signal is input, the teacher data generation unit 114 generates the teacher data TD by omitting the similar normal-time partial data Ndd from among the plurality of normal-time partial data Nd. When the teacher data generation signal is not input, the teacher data generation unit 114 generates the teacher data TD using the plurality of normal-time partial data Nd without omitting the similar normal-time partial data Ndd.

[0053] FIG. 5 is a diagram showing an example of obtaining teacher data by extracting a plurality of normal-time partial data Nd from the normal-time continuous data ND using the sliding window S and omitting the similar normal-time partial data Ndd having continuity before and after. In the example shown in FIG. 5, the similar normal-time partial data Ndd is the data described in italic characters. Also, in the example shown in FIG. 5, the latter half of the normal-time partial data Nd located before the similar normal-time partial data Ndd is the symbol F, and the first half of the normal-time partial data Nd located after the similar normal-time partial data Ndd is the symbol R. In the example shown in FIG. 5, "00", which is the latter half of the normal-time partial data Nd located before the similar normal-time partial data Ndd, is the same as "00", which is the first half of the normal-time partial data Nd located after the similar normal-time partial data Ndd. Therefore, in the example shown in FIG. 5, the similar normal-time partial data Ndd has continuity before and after. Therefore, in the example shown in FIG. 5, the teacher data TD can be generated by omitting the similar normal-time partial data Ndd.

[0054] In this embodiment, when the teacher data generation unit 114 arranges a plurality of normal partial data Nd in the order of generation and can ensure the continuity between the normal partial data Nd located before the similar normal partial data Ndd and the normal partial data Nd located after the similar normal partial data Ndd, the teacher data TD is generated by omitting the similar normal partial data Ndd.

[0055] Thereby, even when the similar normal partial data Ndd determined to be similar by the similarity determination unit 13 is omitted in the plurality of normal partial data Nd, the continuity in the plurality of normal partial data Nd after omission can be ensured. Therefore, it is possible to generate the teacher data TD while ensuring continuity and omitting the similar normal partial data Ndd that are similar.

[0056] Therefore, with the configuration of this embodiment, the teacher data TD can be restored to the original normal continuous data ND. That is, with the configuration of this embodiment, reversible processing of the teacher data TD can be performed. Thereby, after generating the teacher data TD, the specifications (window width, movement amount) of the sliding window S can be changed to adjust the data amount and accuracy of the teacher data TD, and the degree of freedom in generating the teacher data TD can be improved. Moreover, the plurality of normal partial data Nd after the omission can be visualized in a graph or the like.

[0057] Also, in this embodiment, when the rear data F of the normal partial data Nd located before the similar normal partial data Ndd and the front data R of the normal partial data Nd located after the similar normal partial data Ndd match, at least the similar normal partial data Ndd is omitted to generate the teacher data TD.

[0058] Thereby, in the plurality of normal partial data Nd, the similar normal partial data Ndd determined to be similar by the similarity determination unit 13 can be omitted while ensuring the continuity of the normal partial data Nd.

[0059] (Modification example) Next, a modification of Embodiment 2 will be described. In this modification, the continuity confirmation unit 115 confirms the continuity between the normal-time partial data Nd located before the similar normal-time partial data Ndd and the partial data Nd located at a predetermined position behind the similar normal-time partial data Ndd.

[0060] The predetermined position is a position behind the similar normal-time partial data Ndd by the number of normal-time partial data Nd corresponding to the number of data determined by the window width of the sliding window S. For example, when the window width of the sliding window S is 3, since the number of data is 3, the predetermined position is the position of the normal-time partial data Nd located three positions behind the similar normal-time partial data Ndd.

[0061] Specifically, the continuity confirmation unit 115 determines whether the latter half of the normal-time partial data Nd located before the similar normal-time partial data Ndd matches the former half of the normal-time partial data Nd located at a predetermined position behind the similar normal-time partial data Ndd. The method for determining this match, and the definitions of the latter half and the former half are the same as those in Embodiment 2.

[0062] When the continuity confirmation unit 115 determines that the partial data located before the similar normal-time partial data Ndd and the partial data located at a predetermined position behind the similar normal-time partial data Ndd have continuity, a teacher data generation signal is generated to cause the teacher data generation unit 114 to generate teacher data TD by omitting predetermined partial data including the similar normal-time partial data Ndd. The teacher data generation signal is input to the teacher data generation unit 114.

[0063] The teacher data generation unit 114 generates teacher data TD based on the teacher data generation signal output from the continuity confirmation unit 115. Specifically, when the teacher data generation signal is input, the teacher data generation unit 114 generates teacher data TD by omitting the partial data located between the partial data located before the similar normal-time partial data Ndd and the partial data located at a predetermined position behind the similar normal-time partial data Ndd from among the plurality of partial data.

[0064] FIG. 6 is a diagram showing an example of obtaining teacher data TD by extracting a plurality of normal-time partial data Nd from normal-time continuous data ND using a slide window S and omitting the normal-time partial data located between the normal-time partial data Nd located before the similar normal-time partial data Ndd and the normal-time partial data Nd located at a predetermined position after the similar normal-time partial data Ndd when they are continuous. In the example shown in FIG. 6, the similar normal-time partial data Ndd is data described in italic characters. Also, in the example shown in FIG. 6, the latter half of the normal-time partial data Nd located before the similar normal-time partial data Ndd is symbol F, and the former half of the normal-time partial data Nd located at a predetermined position after the similar normal-time partial data Ndd is symbol R. In the example shown in FIG. 6, "12", which is the latter half of the normal-time partial data Nd located before the similar normal-time partial data Ndd, is the same as "12", which is the former half of the normal-time partial data Nd located at a predetermined position after the similar normal-time partial data Ndd. Therefore, in the example shown in FIG. 6, among the plurality of normal-time partial data Nd, the normal-time partial data Nd from the similar normal-time partial data Ndd to before the predetermined position (including the similar normal-time partial data Ndd, the normal-time partial data Nd for the number of data of the slide window S) can be omitted to generate the teacher data TD.

[0065] Note that, in the example shown in FIG. 6, the teacher data TD also includes the similar normal-time partial data Ndd (data described in italic characters in FIG. 6). However, "12", which is the latter half of the normal-time partial data Nd located before the similar normal-time partial data Ndd, is different from "11", which is the former half of the normal-time partial data Nd located at a predetermined position after the similar normal-time partial data Ndd. Therefore, it cannot be omitted in the plurality of normal-time partial data Nd.

[0066] With the above configuration, while ensuring continuity, similar similar normal-time partial data Ndd can be omitted to generate the teacher data TD. Thereby, the teacher data TD can be restored to the original normal-time continuous data ND. That is, with the configuration of this modified example, reversible processing of the teacher data TDd can be performed.

[0067] (Other embodiments) As described above, embodiments of the present invention have been explained. However, the above-described embodiments are merely examples for implementing the present invention. Therefore, without being limited to the above-described embodiments, it is possible to appropriately modify and implement the above-described embodiments within the scope not departing from the gist thereof.

[0068] In each of the above embodiments, the target device P, which is the target of abnormality detection by the abnormality detection devices 1 and 100, is, for example, a vibration conveyor, a vibration generator, or a press. However, the target device may be any device as long as it is a device in which an abnormality occurs.

[0069] In each of the above embodiments, the abnormality detection devices 1 and 100 may have a storage unit that stores at least one of normal continuous data, normal partial data, data related to the result of similarity determination, teacher data, measurement continuous data, measurement partial data, data related to the calculation result of the degree of abnormality, and data related to abnormality detection.

[0070] In each of the above embodiments, in the examples shown in FIGS. 3, 5, and 6, the sliding window S has a length that includes three data. However, each figure is merely an example figure, and the length of the sliding window may be any length, and the moving amount of the sliding window may be any moving amount.

[0071] In the first embodiment, the abnormality detection unit 16 of the abnormality detection device 1 detects an abnormality of the target device P based on the degree of abnormality calculated by the degree of abnormality calculation unit 15. However, the abnormality detection device may not have an abnormality detection unit. The abnormality detection device may output the degree of abnormality calculated by the degree of abnormality calculation unit.

Industrial Applicability

[0072] The present invention can be used for an abnormality detection device that detects an abnormality of a target device.

Explanation of Reference Numerals

[0073] 1. 100 Abnormality Detection Device 11 Normal-Time Continuous Data Acquisition Unit 12 Normal-Time Partial Data Extraction Unit 13 Similarity Judgment Unit 14, 114 Teacher Data Generation Unit 15 Abnormality Degree Calculation Unit 16 Abnormality Detection Unit 115 Continuity Confirmation Unit 21 Measurement Continuous Data Acquisition Unit 22 Measurement Partial Data Extraction Unit P Target Device ND Normal-Time Continuous Data Nd Normal-Time Partial Data Ndd Similar Normal-Time Partial Data TD Teacher Data MD Measurement Continuous Data Md Measurement Partial Data S Slide Window

Claims

1. An abnormality detection device that detects an abnormality of a target device by using teacher data obtained based on normal continuous data related to a physical quantity of the target device and measurement continuous data obtained by measuring the physical quantity of the target device, a normal partial data extraction unit that extracts a plurality of normal partial data from the normal continuous data, a similarity determination unit that determines similar normal partial data among the plurality of normal partial data extracted by the normal partial data extraction unit, a teacher data generation unit that generates the teacher data by omitting the similar normal partial data determined to be similar by the similarity determination unit among the plurality of normal partial data, a measurement partial data extraction unit that extracts a plurality of measurement partial data from the measurement continuous data, an abnormality degree calculation unit that calculates an abnormality degree according to the similarity between the normal partial data included in the teacher data and the measurement partial data, having, abnormality detection device.

2. In the abnormality detection device according to Claim 1, further comprising an abnormality detection unit that detects an abnormality of the target device when the abnormality degree calculated by the abnormality degree calculation unit is equal to or greater than a predetermined value, abnormality detection device.

3. In the abnormality detection device according to Claim 2, the similarity determination unit determines that the two normal partial data are similar when the similarity between the two normal partial data is equal to or less than a threshold value, the threshold value is smaller than the predetermined value, abnormality detection device.

4. In the abnormality detection device according to Claim 1, when the teacher data generation unit can ensure the continuity between the normal partial data located before the similar normal partial data and the normal partial data located after the similar normal partial data in a state where the plurality of normal partial data are arranged in the generation order, the teacher data generation unit omits the similar normal partial data to generate the teacher data, abnormality detection device.

5. In the abnormality detection device according to Claim 4, when the rear data of the normal partial data located before the similar normal partial data and the front data of the normal partial data located after the similar normal partial data match, the teacher data generation unit generates the teacher data by omitting at least the similar normal partial data, abnormality detection device.

Images