Anomaly detection device

The abnormality detection device improves accuracy by using DTW to compare and delete duplicate path information, addressing the accuracy decrease issue in existing devices.

JP7591711B2Active Publication Date: 2024-11-29SINFONIA TECHNOLOGY CO LTD
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
JP2020205202
Authority / Receiving Office
JP · JP
Patent Type
Patents
Current Assignee / Owner
Filing Date
2020-12-10
Publication Date
2024-11-29
Estimated Expiration
2040-12-10

AI Technical Summary

Technical Problem

Existing abnormality detection devices using DTW (dynamic time warping) suffer from decreased accuracy due to duplicate path information, which affects the similarity calculation of waveforms.

Method used

An abnormality detection device that includes comparison means using DTW to compare monitoring data with reference data, path derivation means to derive path information, and duplicate path deletion means to eliminate duplicate path information, thereby improving detection accuracy.

Benefits of technology

The device suppresses the decrease in detection accuracy caused by duplicate path information by effectively deleting such paths, enhancing the efficiency and accuracy of anomaly detection.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 0007591711000005
    Figure 0007591711000005
  • Figure 0007591711000006
    Figure 0007591711000006
  • Figure 0007591711000007
    Figure 0007591711000007
Patent Text Reader

Abstract

To provide an abnormality detecting device capable of restraining a decrease in detection precision because of overlapping of path information.SOLUTION: An abnormality detecting device 1 includes: comparison means 2 for comparing monitoring data as a time series of data for expressing an operation state of equipment for monitoring presence or absence of abnormality with reference data as time series of data for expressing a normal operation state of the equipment; and determination means 3 for determining presence or absence of abnormality on the basis of a comparison result between the monitoring data and the reference data by the comparison means 2. The comparison means 2 includes: path derivation means 21 for deriving plural pieces of path information with a comparison point and a reference point set for the monitoring data as a set by the dynamic time warping method; and overlapped path deletion means 22 for executing extraction processing of overlapped path information in which the comparison point or the reference point is identical from among the plural pieces of path information derived by the path derivation means 21 and deletion processing of the overlapped path information.SELECTED DRAWING: Figure 1
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to an anomaly detection device that detects operational anomalies in equipment such as rotating machines by utilizing DTW (dynamic time warping). [Background technology]

[0002] As an example of the above-mentioned anomaly detection device, the one disclosed in Patent Document 1 is known. In this anomaly detection device, the presence or absence of an anomaly in a device to be monitored for anomalies is detected based on the result of performing DTW on time series data representing the operating state of the device and standard time series data representing the normal operating state of the device. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] JP 2011-203787 A Summary of the Invention [Problem to be solved by the invention]

[0004] Incidentally, the path information output by DTW (hereinafter referred to as path information) may include positions set on the time series data or positions set on the standard time series data that overlap with information on other paths (duplicate path information).

[0005] In this case, the overlapping path information causes an error in the calculation result of the similarity of the waveforms being compared. Therefore, in general devices that use DTW to detect equipment anomalies, including the anomaly detection device described above, there is a problem that the error caused by the overlapping path information reduces the accuracy of anomaly detection.

[0006] In view of the above circumstances, an object of the present invention is to provide an anomaly detection device that can suppress a decrease in detection accuracy due to overlapping of path information. [Means for solving the problem]

[0007] The anomaly detection device of the present invention comprises: A comparison means for comparing monitoring data, which is a time series of data representing the operating state of a device to be monitored for the presence or absence of an abnormality, with reference data, which is a time series of data representing the operating state of the device under normal conditions; a determination means for determining the presence or absence of an abnormality based on a result of the comparison between the monitoring data and the reference data by the comparison means, the comparison means includes a path derivation means for deriving a plurality of path information sets, each set consisting of one of the comparison points and one of the reference points, from among a plurality of comparison points set in the monitoring data and a plurality of reference points set in the reference data, by a dynamic time warping method; The method includes an overlapping path deletion means that executes an extraction process to extract overlapping path information in which at least one of the comparison point and the reference point is identical from the plurality of path information derived by the path derivation means, and a deletion process to delete the overlapping path information.

[0008] According to the anomaly detection device having the above configuration, the comparison means can delete overlapping path information in which at least one of the comparison point and the reference point is identical, and then compare the monitoring data with the reference data, thereby suppressing a decrease in the accuracy of anomaly detection caused by overlapping path information.

[0009] In the anomaly detection device of the present invention, The overlapping path deletion means, in the extraction process, a process of extracting, as first overlapping elements, first elements that are adjacent to each other and have the same comparison point set from among a first array configured with a comparison point included in each of the plurality of path information at the beginning and the end of the first elements; a process of extracting, as second overlapping elements, second elements that are adjacent to each other and have the same reference point set from among a first plurality of second elements and a tail plurality of second elements, for a second array configured with a reference point included in each of the plurality of path information as a second element; and extracting, as the duplicated path, the path information including at least one of the first duplicated element and the second duplicated element.

[0010] According to the anomaly detection device having the above configuration, the efficiency of anomaly detection can be improved by extracting and deleting duplicate path information from the beginning and end of the first and second arrays, where duplicate path information is more likely to occur.

[0011] In the anomaly detection device of the present invention, The overlapping path deletion means, in the extraction process, a process of extracting first overlapping elements, the first overlapping elements being set to the same comparison point as an adjacent first element, from a group of elements from a first element at the beginning of the first array to a first element at the center of the first array and a group of elements from a first element at the end of the first array to a first element at the center of the first array, for a first array configured with a comparison point included in each of the plurality of path information as a first element; a process of extracting second overlapping elements, the second overlapping elements being set to the same comparison point as the adjacent second elements, from a group of elements from a first second element to a center second element of the second array and a group of elements from a last second element to a center first element of the second array, the second overlapping elements being set to the same comparison point as the adjacent second elements ... overlapping elements, the second overlapping elements being set to the same comparison point as the adjacent second overlapping elements, the second overlapping elements being set to the same comparison point as the adjacent second overlapping elements, the second overlapping elements being set to the same comparison point as the adjacent second overlapping elements, and extracting, as the duplicated path, the path information including at least one of the first duplicated element and the second duplicated element.

[0012] According to the anomaly detection device having the above configuration, it is possible to extract and delete overlapping path information including the central parts of the first array and the second array, thereby further improving the accuracy of anomaly detection.

[0013] The anomaly detection device of the present invention comprises: an individual distance derivation means for deriving individual distance information indicating a distance between the first element and the second element for each of the plurality of path information; The vehicle may further include an abnormality determination means for determining the presence or absence of an abnormality based on the individual distance information derived by the individual distance derivation means.

[0014] According to the anomaly detection device having the above configuration, the individual distance between the first and second elements of the path information represents the difference occurring between the comparison point of the monitoring data and the reference point of the reference data, and therefore it becomes possible to detect anomalies based on the individual distance of each path information. Effect of the Invention

[0015] As described above, the anomaly detection device of the present invention can achieve the excellent effect of suppressing a decrease in detection accuracy due to duplication of path information. [Brief description of the drawings]

[0016] [Figure 1] FIG. 1 is a block diagram of an anomaly detection device according to one embodiment of the present invention. [Diagram 2] FIG. 2 is an explanatory diagram of monitoring data of the anomaly detection device according to the embodiment. [Diagram 3] FIG. 3 is an explanatory diagram of reference data of the anomaly detection device according to the embodiment. [Figure 4] FIG. 4 is an explanatory diagram of a warping path of the anomaly detection device according to the embodiment. [Diagram 5] FIG. 5 is an explanatory diagram of index information of the anomaly detection device according to the embodiment. [Figure 6] FIG. 6 is an explanatory diagram of index information before deleting overlapping paths in the anomaly detection device according to the embodiment. [Figure 7] FIG. 7 is an explanatory diagram of index information after the duplicated paths of the anomaly detection device according to the embodiment are deleted. [Figure 8]FIG. 8 is a block diagram of another embodiment of the anomaly detection device of the present invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0017] Hereinafter, an anomaly detection device according to an embodiment of the present invention will be described with reference to the accompanying drawings.

[0018] The anomaly detection device is configured to detect anomalies in equipment such as rotating machines by utilizing DTW (dynamic time warping).

[0019] As shown in FIG. 1, the anomaly detection device 1 of this embodiment includes a comparison means 2 that compares monitoring data, which is a time series of data representing the operating state of an equipment (hereinafter referred to as a monitored equipment) to be monitored for the presence or absence of anomalies, with reference data, which is a time series of data representing the operating state of the equipment under normal conditions, and a determination means 3 that determines the presence or absence of an anomaly based on a comparison result between the monitoring data and the reference data by the comparison means 2.

[0020] The comparison means 2 includes an information acquisition means 20 for acquiring monitoring data and reference data, a path derivation means 21 for deriving, by dynamic time warping, a plurality of path information pairs each consisting of one comparison point and one reference point from a plurality of comparison points set in the monitoring data as inspection points for the abnormality and a plurality of reference points set in the reference data as inspection points for the abnormality, an overlapping path deletion means 22 for executing an extraction process for extracting overlapping path information in which at least one of the comparison point and the reference point is identical from the plurality of path information derived by the path derivation means 21 and a deletion process for deleting the overlapping path information, an individual distance derivation means 23 for deriving individual distance information indicating the distance between the comparison point and the reference point paired for each of the plurality of path information, and a total distance derivation means 24 for deriving total distance information by summing up the individual distances of each path information.

[0021] The information acquisition means 20 may be configured to acquire (read) monitoring data and reference data from a storage device or an external computer, etc., or may be configured to acquire (receive) monitoring data directly from a measuring device such as a sensor.

[0022] The monitoring data is, for example, data obtained by measuring sounds, vibrations, etc. emitted by the monitored device for a predetermined period of time. X is data that can be represented as a waveform on a coordinate system consisting of two mutually perpendicular coordinate axes (coordinate axis H representing time and coordinate axis V representing the measurement value).

[0023] Monitoring Data L X is the comparison point P that is used to check for abnormalities in the waveform. X1、 P X2、 P X3 The process by the path derivation means 21 is performed in a state where a plurality of pieces of comparison point information indicating ... are associated with each other. x and the value V of the measurement axis V x That is, coordinate information indicating the position of the comparison point in the coordinate system is set (see expressions 1 and 2).

number

number

[0024] The reference data is data showing a time series of the normal operating state of the monitored device. As shown in FIG. 3, the reference data L Y is data that can be expressed as a waveform on coordinates consisting of two mutually perpendicular coordinate axes (coordinate axis H representing time and coordinate axis V representing the measurement value).

[0025] Reference Data L Y is the monitoring data L X Reference point P to be used as a reference for comparison with Y1、 P Y2、 P Y3The path derivation means 21 performs processing in a state where a plurality of pieces of reference point information indicating... are associated with each other. Each piece of reference point information includes a value H Y and measurement value axis V Y In other words, the coordinate information indicating the position of the reference point in the coordinate system is set (see expressions 3 and 4).

number

number

[0026] The path derivation means 21 is configured to derive a plurality of pieces of path information as described above. In addition, one piece of path information is a route L connecting a comparison point of the monitoring data and a reference point of the reference data, as shown in FIG. P It is represented as (path).

[0027] As shown in Figure 5, the path derivation means 21 of this embodiment is configured to derive multiple path information (information that combines comparison point information and reference point information) by executing a dynamic time warping method, and to output index information I that combines this multiple path information.

[0028] Therefore, the index information I includes multiple pieces of path information and identification information N for identifying each piece of path information. 1 , N 2 , N 3 ...includes:

[0029] In the index information I of this embodiment, a plurality of pieces of comparison point information form one array (first array), and each piece of comparison point information is an element (first element) that constitutes the first array. Similarly, a plurality of pieces of reference point information form one array (second array), and each piece of reference point information is an element (second element) that constitutes the second array.

[0030] Returning to FIG. 1, the overlapping path deletion means 22 is configured to execute a first search process for searching for overlapping paths based on a first array, and a second search process for searching for overlapping paths based on a second array, in the extraction process.

[0031] The first search process executes a process of extracting, from among a plurality of first elements constituting a first array, first elements that are adjacent to each other and have the same comparison point set thereto, as first overlapping elements.

[0032] In the first search process of this embodiment, a first forward search process is executed to extract a first overlap element from multiple first elements at the beginning of the first array, and a first backward search process is executed to extract a first overlap element from multiple first elements at the end of the first array.

[0033] In the first forward search process, first element designation information for designating a first element to be checked for whether it is a duplicate path is set, and the initial value is set to 0. That is, in the initial value, the first element designation information is set to point to the first element at the beginning.

[0034] In the first forward search process, after selecting the first element indicated by the first element designation information, a selection process is executed in which the first element designation information is advanced by one to further select the first element indicated by the first element designation information (i.e., the first element adjacent to the end of the previously selected first element), and a comparison process is executed in which these two first elements are compared.

[0035] If the comparison points of the two first elements compared in the comparison process are the same, the number set in the first element designation information is set as the first forward cutting position, and the selection process and comparison process are executed again in sequence.

[0036] On the other hand, if the comparison points of the two first elements compared in the comparison process are different, the first element designation information is incremented to the same number as the number of the first array, the first forward search process is terminated, and then the first backward search process is started.

[0037] In the first backward search process, after selecting the first element indicated by the first element designation information, a selection process is executed in which the first element designation information is moved down by one and a first element indicated by the first element designation information (i.e., the first element adjacent to the first element selected earlier at the front side) is further selected, and a comparison process is executed in which these two first elements are compared. Note that, since the first element designation information is moved up to the same number as the number of the first array at the end of the first forward search process, the first element designated in the first backward search process is the last first element.

[0038] If the comparison points of the two first elements compared in the comparison process are the same, the number set in the first element designation information is set as the first rear cut position, and the selection process and comparison process are executed again in sequence.

[0039] On the other hand, if the comparison points of the two first elements compared in the comparison process are different, the first backward search process is terminated.

[0040] The second search process executes a process of extracting, from among a plurality of second elements constituting the second array, second elements that are adjacent to each other and have the same reference point set thereto, as second overlapping elements.

[0041] In the second search process of this embodiment, a second forward search process is executed to extract second duplicate elements from multiple second elements at the beginning of the second array, and a second backward search process is executed to extract second duplicate elements from multiple second elements at the end of the second array.

[0042] In the second forward search process, second element designation information for designating a second element to be checked for whether it is a duplicate path is set, and the initial value is set to 0. That is, in the initial value, the second element designation information is set to point to the first second element.

[0043] In the second forward search process, after selecting the second element indicated by the second element designation information, a selection process is executed in which the second element designation information is advanced by one to further select the second element indicated by the second element designation information (i.e., the second element adjacent to the end of the previously selected second element), and a comparison process is executed in which these two second elements are compared.

[0044] If the reference points of the two second elements compared in the comparison process are the same, the number set in the second element designation information is set as the second forward cutting position, and the selection process and comparison process are executed again in sequence.

[0045] On the other hand, if the reference points of the two second elements compared in the comparison process are different, the second element designation information is incremented to the same number as the number of the second array, the second forward search process is terminated, and then the second backward search process is started.

[0046] In the second backward search process, after selecting the second element indicated by the second element designation information, a selection process is executed in which the second element designation information is moved down by one and a second element indicated by the second element designation information (i.e., the second element adjacent to the first selected second element at the leading side) is further selected, and a comparison process is executed in which these two second elements are compared. Note that, since the second element designation information is moved up to the same number as the number of the second array at the end of the second forward search process, the last second element is the one that is specified first in the second backward search process.

[0047] If the reference points of the two second elements compared in the comparison process are the same, the number set in the second element designation information is set as the second rear cut position, and the selection process and comparison process are executed again in sequence.

[0048] On the other hand, if the reference points of the two second elements compared in the comparison process are different, the second backward search process is terminated.

[0049] The deletion process executes the following processes: a process of comparing the first front cutting position with the second front cutting position and setting the larger one as the front cutting position, a process of comparing the first rear cutting position with the second rear cutting position and setting the smaller one as the rear cutting position, and a process of cutting out the range from the beginning to the position corresponding to the front cutting position and the range from the end to the position corresponding to the rear cutting position in the monitoring data and the reference data. As a result, the range where the comparison point and the reference point where multiple path information are concentrated occurs in the monitoring data and the reference data is cut out. As a result, the index information I becomes in a state where the path information corresponding to the overlapping path is deleted.

[0050] The individual distance deriving means 23 is configured to derive individual distance information for each of the plurality of path information. In addition, the individual distance information becomes larger as the comparison point is farther away from the reference point, so the length of the individual distance information reflects the difference when comparing the comparison point with the reference point.

[0051] The individual distance derivation means 23 may be configured to calculate the distance between the comparison point and the reference point in the direction of another coordinate axis perpendicular to the time axis as the individual distance information. The individual distance derivation means 23 may set the absolute value of the difference between the comparison point and the reference point as the individual distance information, or may set the result of calculation using a formula such as Euclidean distance, Mahalanobis distance, or cosine distance as the individual distance information.

[0052] The total distance calculation means derives total distance information by adding up (for example integrating) the individual distance information of each piece of identification information. The total distance information also indicates the degree of similarity between the monitoring data and the reference data.

[0053] The determination means 3 executes a process of determining whether or not the total distance information is abnormal based on the total distance information. The determination means 3 of this embodiment is configured to use a threshold value as a criterion for determining whether the total distance information is abnormal or normal. Since the total distance information increases as the difference between the monitoring data and the reference data increases, the determination means 3 is configured to determine that the total distance information is abnormal if the total distance information is equal to or greater than the threshold value, and that the total distance information is normal if the total distance information is less than the threshold value.

[0054] The threshold value may be determined based on a result of judgment between normal operation and reference data.

[0055] The configuration of the anomaly detection device 1 according to this embodiment is as described above. Next, the operation of the anomaly detection device 1 will be described.

[0056] When the anomaly detection device 1 is operated, the information acquisition means 20 acquires the monitoring data and the reference data.

[0057] Then, the path derivation means 21 executes dynamic time warping on the monitoring data and the reference data to output index information (see FIG. 6).

[0058] Next, the overlapping path deleting means 22 extracts and deletes overlapping path information from the index information.

[0059] To explain the operation of the overlapping path elimination means 22 using an example of index information, in the index information shown in Figure 6, the comparison point set in the first element at the beginning of the first array is different from the comparison point set in the second first element from the beginning. Therefore, in the first forward search process, the overlapping path elimination means 22 does not move up the first element designation information, but sets the number set in the first element designation information as the first forward cutting position, and terminates the process.

[0060] In addition, in the index information shown in Figure 6, the comparison point set in the first element at the end and the comparison point set in the first element second from the end are different, so in the first forward search process, the overlapping path deletion means 22 sets the number set in the first element designation information to the first backward cutting position without moving the first element designation information down, and terminates the process.

[0061] On the other hand, in the second forward search process for the second array, the reference point set for the first second element and the reference point set for the third second element from the beginning are the same, so after repeating the selection process and comparison process up to the third second element from the beginning, the second forward cut position is set to 3 and the process is terminated.

[0062] In addition, in the second backward search process on the second array, the reference point set in the last second element and the reference point set in the fourth second element from the end are the same, so after repeating the selection process and comparison process up to the fourth second element from the end, the second backward cut position is set to 4 and the process is terminated.

[0063] Then, the first front cutting position is compared with the second front cutting position, and the larger (the number on the tail side) of the second front cutting position, 3, is set as the front cutting position, and the first rear cutting position is compared with the second rear cutting position, and the smaller (the number on the head side) of the second rear cutting position, 4, is set as the rear cutting position, and then the range from the head to the position corresponding to the front cutting position and the range from the tail to the position corresponding to the rear cutting position are cut out from the monitoring data and reference data. This causes the first three path information pieces from the head and the fourth path information pieces from the tail, which are duplicate paths, to be deleted from index information I.

[0064] In this manner, comparison points and reference points where a plurality of pieces of path information are concentrated are cut out from the monitoring data and the reference data, and overlapping path information is deleted.

[0065] Then, the individual distance information is derived for each of the multiple path information, the total distance calculation means derives total distance information, and the judgment means 3 executes a process of judging whether or not there is an abnormality based on the total distance information, thereby completing the series of operations of the anomaly detection device.

[0066] As described above, according to the anomaly detection device 1 of this embodiment, the comparison means 2 can delete overlapping path information in which at least either the comparison point or the reference point is identical, and then compare the monitoring data with the reference data, thereby suppressing a decrease in the accuracy of anomaly detection caused by overlapping path information.

[0067] Furthermore, by extracting and deleting duplicated path information from the beginning and end of the first and second arrays where duplicated path information is likely to occur, it is possible to prevent a decrease in accuracy of anomaly detection.

[0068] The anomaly detection device 1 according to the present invention is not limited to the above embodiment, and it goes without saying that various modifications can be made without departing from the gist of the present invention.

[0069] Although not specifically mentioned in the above embodiment, the anomaly detection device may include a detailed anomaly detection means 4 for determining the presence or absence of an anomaly based on the individual distance information of each path information, as shown in FIG.

[0070] In this way, since the determination means 3 determines an abnormality based on the total distance information, it may be difficult to detect an abnormality that occurs locally in the monitoring data, but since the abnormality determination means 3 detects an abnormality based on the individual distance information in the state before it is integrated into the total distance information, it becomes possible to detect an abnormality that occurs locally in the monitoring data. Note that the abnormality detection process by the detailed abnormality detection means 4 may be executed before the abnormality detection process by the determination means 3.

[0071] Although not specifically mentioned in the above embodiment, in the deletion process, after cutting out the monitoring data and reference data based on the front cutting position and the rear cutting position, the beginning and end of the monitoring data and reference data may be further cut out.

[0072] In the above embodiment, the first forward search process always starts processing in order from the first element at the top, but is not limited to this configuration. For example, the first overlapping elements may be extracted in order from the first element at the top, and the first forward search process may start when the number of extracted first overlapping elements reaches a predetermined number. In this case, the initial value of the number of extractions is set to 1, and if the first element at the top and the second first element from the top are not first overlapping elements, the number of extractions is set to a number of 5% to 10% of the entire first array.

[0073] Similarly, the first backward search process always starts from the last first element in order, but is not limited to this configuration. For example, the first duplicate elements may be extracted in order from the last first element, and the first backward search process may start when the number of extracted first duplicate elements reaches a predetermined number. In this case, the initial value of the number of extractions is set to the total number of first elements, and even if the last first element and the second last first element are not first duplicate elements, the number of extractions is set to a number of 5% to 10% of the entire first array.

[0074] In this case, it is preferable that the first forward search process and the first backward search process are configured to forcibly terminate the process when the first element in the center is reached in both the first forward search process and the first backward search process. In this way, it is possible to prevent the first forward cut position from being set to the end of the first element in the first forward search process, and the first backward cut position from being set to the start of the first element in the first backward search process, i.e., to prevent most of the path information from being deleted.

[0075] In the above embodiment, the second forward search process always starts from the first second element in order, but is not limited to this configuration. For example, the second duplicate elements may be extracted in order from the first second element, and the second forward search process may start when the number of extracted second duplicate elements reaches a predetermined number. In this case, the initial value of the number of extractions is set to 1, and if the first second element and the second second element from the top are not second duplicate elements, the number of extractions is set to 5% to 10% of the entire second array.

[0076] Similarly, the second backward search process always starts from the last second element in order, but is not limited to this configuration. For example, the second duplicate elements may be extracted in order from the last second element, and the second backward search process may start when the number of extracted second duplicate elements reaches a predetermined number. In this case, the initial value of the number of extractions is set to the total number of second elements, and even if the last second element and the second last second element are not second duplicate elements, the number of extractions is set to a number of 5% to 10% of the entire second array.

[0077] It is preferable that the second forward search process and the second backward search process are configured to forcibly terminate the process when the second element in the center is reached in both the second forward search process and the second backward search process. In this way, it is possible to prevent the second forward cut position from being set to the end of the second element in the second forward search process and the second backward cut position from being set to the start of the second element in the second backward search process, i.e., to prevent most of the path information from being deleted. [Explanation of symbols]

[0078] 1... anomaly detection device, 2... comparison means, 3... determination means, 20... information acquisition means, 21... path derivation means, 22... duplicate path deletion means, 23... individual distance derivation means, 24... total distance derivation means

Claims

1. A comparison means for comparing monitoring data, which is a time series of data representing the operating state of a device to be monitored for anomalies and is expressed as a waveform, with reference data, which is a time series of data representing the operating state of the device under normal conditions and is expressed as a waveform; a determination means for determining the presence or absence of an abnormality based on a result of the comparison between the monitoring data and the reference data by the comparison means, the comparison means includes a path derivation means for deriving a plurality of path information sets, each set consisting of one of the comparison points and one of the reference points, from among a plurality of comparison points set in the monitoring data and a plurality of reference points set in the reference data, by a dynamic time warping method; an overlapping path deletion means for executing an extraction process for extracting overlapping path information in which at least one of the comparison point and the reference point is the same from the plurality of path information derived by the path derivation means, and a deletion process for deleting the overlapping path information, The overlapping path deletion means, in the extraction process, a process of extracting, as first overlapping elements, first elements that are adjacent to each other and have the same comparison point set from among a plurality of first elements at the beginning and a plurality of first elements at the end of a first array configured with the comparison points included in each of the plurality of path information as first elements; a process of extracting, as second overlapping elements, second elements adjacent to each other and having the same reference point set from among a first plurality of second elements and a tail plurality of second elements, for a second array configured with the reference points included in each of the plurality of path information as second elements; extracting, as the overlapping path information, the path information including at least one of the first overlapping element and the second overlapping element. Anomaly detection device.

2. A comparison means for comparing monitoring data, which is a time series of data representing the operating state of a device to be monitored for anomalies and is expressed as a waveform, with reference data, which is a time series of data representing the operating state of the device under normal conditions and is expressed as a waveform; a determination means for determining the presence or absence of an abnormality based on a result of the comparison between the monitoring data and the reference data by the comparison means, the comparison means includes a path derivation means for deriving a plurality of path information sets, each set consisting of one of the comparison points and one of the reference points, from among a plurality of comparison points set in the monitoring data and a plurality of reference points set in the reference data, by a dynamic time warping method; an overlapping path deletion means for executing an extraction process for extracting overlapping path information in which at least one of the comparison point and the reference point is the same from the plurality of path information derived by the path derivation means, and a deletion process for deleting the overlapping path information, The overlapping path deletion means, in the extraction process, a process of extracting first overlapping elements, the comparison point of which is the same as that of an adjacent first element, from a group of elements from a first element at the beginning of the first array to a first element at the center of the first array and a group of elements from a first element at the end of the first array to a first element at the center of the first array, for a first array configured with the comparison point included in each of the plurality of path information as a first element; a process of extracting second overlapping elements, the same as the reference point as adjacent second elements, from a group of elements from a first second element to a center second element of the second array and a group of elements from a last second element to a center second element of the second array, the second overlapping elements being configured with the same reference point as adjacent second elements; extracting, as the overlapping path information, the path information including at least one of the first overlapping element and the second overlapping element. Anomaly detection device.

3. an individual distance derivation means for deriving individual distance information indicating a distance between the second elements for each of the plurality of path information; and an abnormality determination means for determining the presence or absence of an abnormality based on the individual distance information derived by the individual distance derivation means. The abnormality detection device according to claim 1 or 2.

Citation Information

Patent Citations

  • Standard time-series data calculation method, abnormality detection method, standard time-series data calculation device and abnormality detection device

    JP2009157787A

  • Data-monitoring method and data-monitoring apparatus

    JP2011203787A

  • Time axis expansion / contraction device, batch process monitoring device, time axis expansion / contraction system and program

    JP2018032319A

  • Operation analysis device, operation analysis method, operation analysis program, and operation analysis system

    JP2020119198A