KR102553397B1 - data processing unit - Google Patents

Abstract

The data processing device 100 includes a time-series acquisition unit 110 that acquires time-series data, and a feature extraction unit 120 that extracts feature quantities of the time-series data based on the time-series data acquired by the time-series data acquisition unit 110. And, based on the feature amount extracted by the feature extractor 120, a first division unit 130 for dividing the time-series data into a plurality of first partial time-series data, and a feature amount extracted by the feature extraction unit 120 based on, a second dividing unit 140 for dividing each of the plurality of first partial time-series data divided by the first dividing unit 130 into a plurality of second partial time-series data.

Description

data processing unit

The present disclosure relates to a data processing apparatus.

A technique for dividing time series data into a plurality of partial time series data is known.

For example, in Patent Literature 1, an extraction condition input unit that accepts input of waveform data including a change point of the device state, parameter information of the waveform data, and device transition information, time-series data of the device, and waveform data A similarity calculation unit that calculates the similarity of the device, an operation mode determination unit that sets the device state based on the transition information of the device, and a change from the time-series data of the device based on the calculated similarity and the determined device state. A change point detection unit that detects a point and sets the start time and end time of a segment, which is a partial sequence of time series data, and an information output unit that outputs the state of the device, the start time of a segment, and the end time of a segment as segment information. A data processing device provided is disclosed.

The data processing device described in Patent Literature 1 (hereinafter referred to as "conventional data processing device") is used to detect a waveform pattern from time-series data using previously prepared waveform data serving as indicators for segmentation and parameter information of the waveform data. , the time-series data is divided into segments that are partial time-series data for each waveform data that repeatedly appears in the time-series data.

[Patent Document 1] International Publication No. 2020/008533

In a conventional data processing apparatus, in order to divide time-series data into segments that are partial time-series data, it is necessary to prepare in advance waveform data serving as an index for division and parameter information (hereinafter referred to as "special information") of the waveform data. There was a problem that there was.

The present disclosure is intended to solve the above-mentioned problems, and provides a data processing device capable of dividing time-series data into partial time-series data for each waveform pattern that repeatedly appears in time-series data without preparing special information in advance. is aiming for

A data processing apparatus according to the present disclosure includes: a time-series acquisition unit that acquires time-series data; a feature extraction unit that extracts a feature amount of time-series data based on the time-series data acquired by the time-series data acquisition unit; and a feature amount extracted by the feature extraction unit. A first division unit for dividing the time-series data into a plurality of first partial time-series data based on , and each of the plurality of first partial time-series data divided by the first division unit based on the feature amount extracted by the feature extraction unit. A second division unit for dividing into a plurality of second partial time-series data, and time-series data acquired by the time-series acquisition unit based on at least one second partial time-series data among the plurality of second partial time-series data divided by the second division unit. An abnormality detection unit is provided to detect anomaly, each of a plurality of first partial time-series data divided by the first division unit has first oscillation interval time-series data and first stable interval time-series data, and the second division unit comprises the first For each of the plurality of first partial time-series data divided by the dividing unit, the first oscillation section time-series data among the first partial time-series data is divided into a plurality of second partial time-series data; Each of the partial time series data has second oscillation interval time series data and second stable interval time series data. , The time-series acquisition unit detects anomalies in the acquired time-series data.

According to the present disclosure, time-series data can be divided into partial time-series data for each waveform pattern that repeatedly appears in time-series data without preparing special information in advance.

1 is a block diagram showing an example of the configuration of a main part of a data processing device according to Embodiment 1 and a data processing system to which the data processing device is applied.
2 is an explanatory diagram showing an example of time-series data acquired by a time-series acquisition unit included in the data processing device according to Embodiment 1, and first partial time-series data obtained by dividing the time-series data by a first division unit.
3 shows first partial time-series data obtained by dividing a first division unit included in the data processing device according to Embodiment 1, and second partial time-series data obtained by dividing the first partial time-series data by a second division unit. It is explanatory drawing which shows an example.
4(a) and 4(b) are diagrams showing an example of the hardware configuration of main parts of the data processing apparatus according to the first embodiment.
5 is a flowchart for explaining an example of processing of the data processing device according to the first embodiment.

Hereinafter, this specified embodiment will be described in detail with reference to the drawings.

(Embodiment 1)

Referring to FIGS. 1 to 5 , the data processing device 100 according to the first embodiment and the data processing system 1 to which the data processing device 100 is applied will be described.

1 is a block diagram showing an example of the configuration of a main part of a data processing apparatus 100 according to Embodiment 1 and a data processing system 1 to which the data processing apparatus 100 is applied.

A data processing system 1 according to Embodiment 1 includes a storage device 10 , a display device 20 , and a data processing device 100 .

The data processing apparatus 100 acquires time-series data and performs predetermined data processing on the acquired time-series data. Details of the data processing device 100 will be described later.

The storage device 10 is a device that stores information necessary for the data processing device 100 to perform data processing. For example, the storage device 10 stores time series data in advance.

The display device 20 is a device such as a display that displays and outputs a display image indicated by display information output from the data processing device 100 . That is, the data processing device 100 outputs the display information to the display device 20 and causes the display device 20 to display a display image indicated by the display information.

The data processing device 100 according to Embodiment 1 includes a time series acquisition unit 110, a feature extraction unit 120, a first division unit 130, a second division unit 140, an anomaly detection unit 150, and an output unit 190.

In addition, the abnormality detection unit 150 is not an essential component of the data processing device 100 .

In Embodiment 1, the data processing apparatus 100 is described as having the abnormality detection unit 150.

The time series acquisition unit 110 acquires time series data.

The time-series data acquired by the time-series acquisition unit 110 is time-series information indicating physical quantities measured, measured, observed, or counted at predetermined time intervals.

Specifically, the time-series data acquired by the time-series acquisition unit 110 is obtained by converting a signal output from a sensor such as a vibration sensor, a distance sensor, a rotation sensor, a gyro sensor, a temperature sensor, or a sound sensor into time-series information. will be. Time-series data is not limited to information obtained by converting a signal output from a sensor into time-series information, as long as it is time-series information representing a physical quantity measured, measured, observed, or counted at predetermined time intervals. In addition, the predetermined time interval does not have to be a uniform interval, and the time interval includes an arbitrary interval.

Specifically, for example, the time-series acquisition unit 110 acquires the time-series data by reading the time-series data previously stored in the storage device 10 .

Since the time series acquisition unit 110 only needs to be capable of acquiring time series data, the acquisition source of the time series data acquired by the time series acquisition unit 110 or the method for acquiring the time series data by the time series acquisition unit 110 are as follows: Not limited.

Hereinafter, in Embodiment 1, the time-series data acquired by the time-series acquisition unit 110 is described as one obtained by converting a signal output from a vibration sensor installed in a processing device into time-series information.

The feature extraction unit 120 extracts feature values of the time-series data based on the time-series data acquired by the time-series acquisition unit 110 .

Specifically, for example, the feature extraction unit 120 extracts a feature amount from the time-series data acquired by the time-series acquisition unit 110 using a slide window. That is, the feature amount output by the feature extractor 120 is information in which a value representing a feature of a region in time series data corresponding to a slide window is presented in time series.

More specifically, for example, the feature extraction unit 120 uses a slide window to extract a value obtained by subtracting the minimum value from the maximum value of the time-series data in the slide window (hereinafter referred to as "amplitude value") as a feature amount. . The feature amount output by the feature extraction unit 120 is information in which the amplitude value of the region in the time series data corresponding to the slide window is expressed in time series.

Since the method of extracting a feature amount from time-series data using a slide window is well-known, description thereof is omitted.

The first division unit 130 divides the time-series data into a plurality of first partial time-series data based on the feature amount extracted by the feature extraction unit 120 .

For example, the feature extraction unit 120 extracts a first feature from the time-series data acquired by the time-series acquisition unit 110 using a first slide window having a predetermined first time length. The first division unit 130 divides the time-series data into a plurality of first partial time-series data based on the first feature extracted by the feature extraction unit 120 .

Specifically, for example, the first division unit 130 compares the amplitude value indicated by the first characteristic quantity with a predetermined threshold value based on the first characteristic quantity, and the amplitude value below the threshold value is greater than the threshold value. The position in the first feature that increases is specified. In the first division 130, the end of the slide window corresponding to the position is a change point of the time-series data (hereinafter referred to as a “first change point”), and the time-series data at the first change point. By dividing the , the time series data is divided into a plurality of first partial time series data.

For example, the first division unit 130 generates first division information indicating division positions in the time-series data of each first partial time-series data divided by the first division unit 130 .

In addition, the first division unit 130 specifies the position in the first feature value at which the amplitude value greater than the threshold value is equal to or less than the threshold value. In the first division 130, the initial stage of the slide window corresponding to the position is a change point of the time series data (hereinafter referred to as a “second change point”), and each first partial time series at the second change point By dividing the data, each of the plurality of first partial time-series data divided by the first dividing unit 130 may be divided into first vibration interval time-series data and first stable interval time-series data.

When the first divider 130 divides each of the plurality of first partial time-series data into first oscillation interval time-series data and first stable interval time-series data, the first divider 130 divides the plurality of divided time-series data. Each of the first partial time-series data of has a first oscillation interval time-series data and a first stable interval time-series data.

In this case, for example, the first division unit 130, for each first partial time series data divided by the first division unit 130, the time series of the first oscillation interval time series data and the first stable interval time series data First division information indicating a division position in data is generated.

In this case, the threshold value is a predetermined value in order to divide the first partial time-series data into the first vibration interval time-series data and the first stable interval time-series data, as described above. Therefore, for example, when the time-series data acquired by the time-series acquisition unit 110 is one obtained by converting a signal output from a sensor such as a vibration sensor into time-series information, the threshold value is set to 1 times the resolution of the sensor. The first stable interval time series data can be extracted from the first partial time series data by determining with about twice the precision in . In addition, the same threshold value can be used with the same value if the device to be measured, such as a sensor or a processing device equipped with a sensor, has the same specifications. Therefore, the said threshold value can be determined easily.

The second division unit 140 converts each of the plurality of first partial time series data divided by the first division unit 130 into a plurality of second partial time series data based on the feature amount extracted by the feature extraction unit 120. split into data

For example, the feature extraction unit 120 uses a second slide window having a predetermined second time length that is shorter than the first time length, and the plurality of first partial time series data divided by the first division unit 130. A second feature is extracted from each of The second division unit 140 divides the first partial time-series data into a plurality of second partial time-series data based on the extracted second feature.

Specifically, for example, the second division unit 140 compares the amplitude value indicated by the second characteristic quantity with a predetermined threshold value based on the second characteristic quantity, and the amplitude value below the threshold value is greater than the threshold value. The position in the second feature that increases is specified. In the second division 140, the end of the slide window corresponding to the position is a change point of the first partial time series data (hereinafter referred to as a "third change point"), and the time series data at the third change point. By dividing , the first partial time-series data is divided into a plurality of second partial time-series data.

For example, for each of the plurality of first partial time-series data, the second dividing unit 140 determines the division position in the time-series data of each second partial time-series data divided by the second dividing unit 140. Second division information indicating

Further, the second division unit 140 specifies the position in the second feature value at which the amplitude value greater than the threshold value is equal to or less than the threshold value. In the second division unit 140, the initial stage of the slide window corresponding to the position is a change point of the time series data (hereinafter referred to as a "fourth change point"), and each second partial time series at the fourth change point. By dividing the data, each of the plurality of second partial time-series data divided by the second dividing unit 140 may be divided into second oscillation interval time-series data and second stable interval time-series data.

When the second divider 140 divides each of the plurality of second partial time-series data into second oscillation interval time-series data and second stable interval time-series data, the second divider 140 divides the plurality of divided time-series data. Each of the second partial time-series data of has second oscillation interval time-series data and second stable interval time-series data.

In this case, for example, the second division unit 140, for each of the second partial time series data divided by the second division unit 140, the time series of the second oscillation interval time series data and the second stable interval time series data Second division information indicating the division position in the data is generated.

In this case, the threshold value is a predetermined value in order to divide the second partial time-series data into the second vibration interval time-series data and the second stable interval time-series data, as described above. Therefore, for example, when the time-series data acquired by the time-series acquisition unit 110 is one obtained by converting a signal output from a sensor such as a vibration sensor into time-series information, the threshold value is set to 1 times the resolution of the sensor. The second stable interval time-series data can be extracted from the second partial time-series data by determining with about twice the precision in . In addition, the same threshold value can be used with the same value if the device to be measured, such as a sensor or a processing device equipped with a sensor, has the same specifications. Therefore, the said threshold value can be determined easily.

Referring to FIG. 2 , the time-series data acquired by the time-series acquisition unit 110 of the data processing device 100 according to the first embodiment, and the second obtained by dividing the time-series data by the first division unit 130 Part 1 describes time series data.

2 shows time-series data acquired by the time-series acquisition unit 110 included in the data processing device 100 according to Embodiment 1, and a first portion obtained by dividing the time-series data by the first division unit 130. It is an explanatory diagram showing an example of time series data.

When the time-series data acquired by the time-series acquisition unit 110 is a signal output from a vibration sensor installed in a processing device that repeatedly manufactures the same processed product, converted into time-series information, the time-series data includes a processed product. The transition of the same time-series value appears repeatedly whenever it manufactures.

The first division unit 130 divides the time-series data into a plurality of first partial time-series data so that the time-series data corresponds to a period during which each of the plurality of processed products is manufactured (hereinafter referred to as "manufacturing period"). It can be divided into a first partial time series data that

In the manufacturing period, a processing period in which a processed member, which is a member to be processed, is processed into a processed product, and an idling period between the time when processing of an arbitrary processed product is finished and the time when processing of the next processed product is started period exists.

The first dividing unit 130 divides the first partial time series data into first oscillation interval time series data and first stable interval time series data, thereby dividing the first partial time series data into a first oscillation interval corresponding to the processing period. It may be divided into time series data and first stable interval time series data corresponding to the idling period.

Referring to FIG. 3 , the first partial time-series data obtained by dividing by the first dividing unit 130 provided in the data processing device 100 according to the first embodiment, and the second dividing unit 140 are the first The second partial time series data obtained by dividing the partial time series data will be described.

3 shows first partial time-series data obtained by division by the first division unit 130 provided in the data processing device 100 according to Embodiment 1, and first partial time-series data obtained by the second division unit 140 It is an explanatory diagram showing an example of the second partial time series data obtained by dividing the data.

When the time-series data acquired by the time-series acquisition unit 110 is a signal output from a vibration sensor installed in a processing device that repeatedly manufactures the same processed product, converted into time-series information, the first partial time-series data includes: In one processing period, a transition of time series values corresponding to each of a plurality of processing steps appears.

Here, the machining process is a process of installing a processing member on a cradle, a process of detecting the shape of a processing member installed on a cradle, a process of cutting a processing member installed on a cradle, a process of inspecting the shape of a processing member after cutting, or, It is a process of removing the processed member after cutting from the cradle, and the like.

The second dividing unit 140 divides each of the first oscillation interval time-series data in the plurality of first partial time-series data into second partial time-series data, so that the first oscillation interval time-series data is processed into a plurality of pieces. It can be divided into second partial time-series data corresponding to periods corresponding to each process (hereinafter referred to as “process period”).

In each of the plurality of process periods, a process operation period in which the processing device actually operates, and a time from when the operation of the processing device ends in a certain processing process to the time when the operation of the processing device starts in the next processing process There is a process idling period between

The second divider 140 divides the second partial time series data into second vibration interval time series data and second stable interval time series data, thereby dividing the second partial time series data into second vibration intervals corresponding to the process operation period. It may be divided into interval time series data and second stable interval time series data corresponding to the process idling period.

With the configuration as described above, the data processing apparatus 100 can divide the time-series data into partial time-series data for each waveform pattern that repeatedly appears in the time-series data, if the first time length and the second time length are determined in advance. there is.

In a conventional data processing apparatus, a user analyzes time-series data obtained in the past and prepares waveform data serving as an index for segmenting the time-series data and parameter information of the waveform data (hereinafter referred to as "special information") in advance. There was a need. Analysis of time series data requires considerable knowledge, and it is not easy for users to prepare special information in advance.

In contrast, for example, when the time-series data acquired by the time-series acquisition unit 110 is a signal output from a vibration sensor installed in a processing device that repeatedly manufactures the same processed product, converted into time-series information. , since both the idling period and the process idling period are known, the user can easily determine the first length of time and the second length of time based on the idling period and the process idling period.

As a result, the data processing apparatus 100 can divide the time-series data into partial time-series data for each waveform pattern that repeatedly appears in the time-series data without preparing special information in advance.

The output unit 190 outputs the first division information and the second division information based on the first division information generated by the first division unit 130 and the second division information generated by the second division unit 140, for example. A display image for visualizing the second division information is generated, and display information indicating the display image is output.

Specifically, for example, the output unit 190 outputs display information to the display device 20 and causes the display device 20 to display a display image indicated by the display information.

The output unit 190 is not limited to outputting display information. For example, the output unit 190 may output the first division information and the second division information to the storage device 10 and store the first division information and the second division information in the storage device 10. good night.

As described above, the data processing device 100 may include the abnormality detection unit 150 .

The anomaly detecting unit 150 is anomaly in the time-series data acquired by the time-series acquisition unit 110 based on at least one piece of second partial time-series data among a plurality of second partial time-series data divided by the second division unit 140. detect For example, the abnormality detection unit 150 generates detection information indicating a detection result.

Specifically, for example, based on the second stable interval time-series data among the plurality of second partial time-series data divided by the second division unit 140, the time-series data acquired by the time-series acquisition unit 110 is abnormal. detect

More specifically, for example, the anomaly detection unit 150 determines that the number of second stable interval time-series data corresponding to each of the plurality of first partial time-series data divided by the first division unit 130 is all first In the partial time-series data, abnormality of the time-series data is detected by determining whether or not they are the same.

If the number of second stable interval time series data corresponding to each of the plurality of first partial time series data is not the same in all first partial time series data, for example, the period between the two processing steps , it is considered that there is a possibility that the period is shorter than the predetermined period. Further, in this case, it is conceivable that there is a possibility that unexpected vibration occurs in the processing device for some reason during the period between the two processing steps.

Therefore, by configuring as described above, the anomaly detection unit 150 can accurately detect anomalies in time-series data.

Further, for example, the abnormality detection unit 150 determines that the second stable interval time-series data among the plurality of second partial time-series data divided by the second division unit 140 and the second stable interval time-series data are first oscillated. Abnormality in the time-series data acquired by the time-series acquisition unit 110 may be detected based on second stable interval positional information indicating which second stable interval time-series data in the interval time-series data.

For example, the anomaly detection unit 150 inputs second stable interval time-series data and second stable interval positional information to a pre-prepared trained model. The anomaly detection unit 150 detects anomalies in the time-series data based on the inference result output from the learned model. For example, the learned model is previously stored in the storage device 10, and the abnormality detection unit 150 acquires the learned model by reading the learned model from the storage device 10.

By configuring as above, the anomaly detection unit 150 can accurately detect anomalies in time-series data.

In addition, in the learned model, a learning device (not shown) performs supervised learning or unsupervised learning on a pre-prepared learning model based on the second stable interval time series data and the second stable interval position information. created by Description of the method of generating the trained model is omitted.

Further, for example, the anomaly detection unit 150, based on the second stable interval time-series data and the second stable interval position information, converts the second stable interval time-series data into a different first As second stable interval time-series data among a plurality of second partial time-series data obtained by dividing the first oscillation interval time-series data of the partial time-series data, by comparing with second stable interval time-series data having the same second stable interval position information, Abnormalities in the time-series data acquired by the acquisition unit 110 may be detected.

Specifically, for example, the anomaly detection unit 150 compares the second stable interval time series data with another second stable interval time series data having the same second stable interval positional information, and determines the degree of similarity, thereby acquiring the time series. Abnormality of time-series data acquired by unit 110 is detected. Since the method of determining the degree of similarity between time-series data is well-known, description is abbreviate|omitted.

By configuring as above, the anomaly detection unit 150 can accurately detect anomalies in time-series data.

The anomaly detection unit 150 may, when detecting an anomaly in the time-series data, calculate an anomaly degree of the time-series data, and generate detection information indicating the calculated anomaly degree.

By the abnormality detection unit 150 calculating the degree of abnormality of the time-series data, the data processing device 100 determines, for example, an abnormality such as deterioration of the processing device before an abnormality occurs, such as before a processing device fails. signs can be identified.

When the data processing device 100 includes the abnormality detection unit 150, the output unit 190 outputs, for example, detection generated by the abnormality detection unit 150 in addition to the first divided information and the second divided information. A display image for visualizing information is generated, and display information representing the display image is output.

The output unit 190 outputs detection information to the storage device 10 in addition to the first division information and the second division information, and outputs the first division information, the second division information, and the detection information to the storage device 10. It may be to memorize information.

The hardware configuration of main parts of the data processing apparatus 100 according to the first embodiment will be described with reference to Figs. 4(a) and 4(b).

4(a) and 4(b) are diagrams showing an example of the hardware configuration of main parts of the data processing apparatus 100 according to the first embodiment.

As shown in Fig. 4(a), the data processing apparatus 100 is constituted by a computer, and the computer has a processor 401 and a memory 402.

In the memory 402, the computer includes a time series acquisition unit 110, a feature extraction unit 120, a first division unit 130, a second division unit 140, anomaly detection unit 150, and an output unit ( 190), a program for functioning is stored. When the processor 401 reads and executes the program stored in the memory 402, the time series acquisition unit 110, the feature extraction unit 120, the first division unit 130, and the second division unit 140 ), the abnormality detection unit 150, and the output unit 190 are realized.

Also, as shown in FIG. 4( b ), the data processing device 100 may be configured by the processing circuit 403 . In this case, the functions of the time series acquisition unit 110, feature extraction unit 120, first division unit 130, second division unit 140, anomaly detection unit 150, and output unit 190 are processing circuits. This may be realized by (403).

Also, the data processing device 100 may be configured by a processor 401, a memory 402, and a processing circuit 403 (not shown). In this case, some of the functions of the time series acquisition unit 110, feature extraction unit 120, first division unit 130, second division unit 140, anomaly detection unit 150, and output unit 190 The functions may be realized by the processor 401 and the memory 402, and the remaining functions may be realized by the processing circuit 403.

The processor 401 uses, for example, a central processing unit (CPU), a graphics processing unit (GPU), a microprocessor, a microcontroller, or a digital signal processor (DSP).

The memory 402 uses, for example, a semiconductor memory or a magnetic disk. More specifically, the memory 402 includes RAM (Random Access Memory), ROM (Read Only Memory), flash memory, EPROM (Erasable Programmable Read Only Memory), EEPROM (Electrically Erasable Programmable Read-Only Memory), SSD ( Solid State Drive) or HDD (Hard Disk Drive).

The processing circuit 403 is, for example, an Application Specific Integrated Circuit (ASIC), a Programmable Logic Device (PLD), a Field-Programmable Gate Array (FPGA), a System-on-a-Chip (SoC), or a system LSI ( Large-Scale Integration) was used.

Referring to Fig. 5, the operation of the data processing apparatus 100 according to Embodiment 1 will be described.

5 is a flowchart for explaining an example of processing performed by the data processing apparatus 100 according to the first embodiment.

First, in step ST501, the time series acquisition unit 110 acquires time series data.

Next, in step ST502, the feature extraction unit 120 extracts the first feature amount.

Next, in step ST503, the first division unit 130 divides the time-series data into a plurality of first partial time-series data.

Next, in step ST504, the feature extraction unit 120 extracts the second feature.

Next, in step ST505, the second dividing unit 140 divides each of the plurality of first partial time-series data divided by the first dividing unit 130 into a plurality of second partial time-series data.

Next, in step ST506, the abnormality detection unit 150 detects an abnormality in the time series data.

Next, in step ST507, the output unit 190 outputs display information.

After step ST507, the data processing device 100 ends the processing of the flowchart.

As described above, the data processing device 100 includes a time-series acquisition unit 110 that acquires time-series data, and feature extraction that extracts feature quantities of the time-series data based on the time-series data acquired by the time-series data acquisition unit 110. The unit 120, a first division unit 130 for dividing the time-series data into a plurality of first partial time-series data based on the feature amount extracted by the feature extraction unit 120, and the feature extraction unit 120 A second dividing unit 140 for dividing each of the plurality of first partial time-series data divided by the first dividing unit 130 into a plurality of second partial time-series data based on the feature amount to be extracted is provided.

With this configuration, the data processing apparatus 100 can divide the time-series data into partial time-series data for each waveform pattern that repeatedly appears in the time-series data without preparing special information in advance.

In the data processing device 100, in the above-described configuration, the feature extraction unit 120 uses the first slide window having a first predetermined time length to obtain time-series data obtained by the time-series acquisition unit 110. A first feature is extracted from, and the first division unit 130 divides the time series data into a plurality of first partial time series data based on the first feature extracted by the feature extraction unit 120, and the feature extraction unit (120) is a second feature value from each of the plurality of first partial time-series data divided by the first dividing unit 130 using a second slide window having a predetermined second time length shorter than the first time length. is extracted, and the second dividing unit 140 is configured to divide the first partial time-series data into a plurality of second partial time-series data based on the extracted second feature.

With this configuration, the data processing apparatus 100 can divide the time-series data into partial time-series data for each waveform pattern that repeatedly appears in the time-series data without preparing special information in advance.

Further, in addition to the configuration described above, the data processing device 100, based on at least one piece of second partial time-series data among a plurality of second partial time-series data divided by the second division unit 140, a time-series acquisition unit, An anomaly detection unit 150 for detecting anomaly in the time-series data acquired by (110) is provided.

With this configuration, the data processing device 100 can accurately detect anomalies in time-series data.

Further, in the data processing device 100, in the above configuration, each of the plurality of first partial time-series data divided by the first division unit 130 includes the first oscillation interval time-series data and the first stable interval time-series data. And the second dividing unit 140, for each of the plurality of first partial time series data divided by the first dividing unit 130, converts the first oscillation section time series data among the first partial time series data into a plurality of first oscillation interval time series data. Each of the plurality of second partial time-series data divided into two-part time-series data and divided by the second dividing unit 140 has second vibration interval time-series data and second stable interval time-series data, and anomaly detection unit 150 is configured to detect anomalies in the time-series data acquired by the time-series acquisition unit 110 based on the second stable interval time-series data among the plurality of second partial time-series data divided by the second division unit 140.

With this configuration, the data processing device 100 can accurately detect anomalies in time-series data.

Further, in the data processing device 100 in the above-described configuration, the anomaly detection unit 150 includes the second stable interval time-series data among the plurality of second partial time-series data divided by the second division unit 140, and the corresponding time-series data. The time-series data acquired by the time-series acquisition unit 110 based on the second stable interval positional information indicating at what number the second stable interval time-series data is the second stable interval time-series data in the first oscillation interval time-series data. configured to detect abnormalities.

With this configuration, the data processing device 100 can accurately detect anomalies in time-series data.

Further, in the configuration described above, in the data processing device 100, the anomaly detection unit 150 generates the second stable interval time-series data based on the second stable interval time-series data and the second stable interval position information. As second stable interval time series data among a plurality of second partial time series data obtained by dividing the first oscillation interval time series data of the other first partial time series data by the second division unit 140, the second stable interval location information is the same. It is structured so that abnormality of the time-series data acquired by the time-series acquisition unit 110 may be detected by comparing with the interval time-series data.

With this configuration, the data processing device 100 can accurately detect anomalies in time-series data.

Also, in the present disclosure, within the specified scope, modification of any constituent elements of the embodiments or omission of arbitrary constituent elements in the embodiments is possible.

A data processing apparatus according to the present disclosure can be applied to a data processing system that detects abnormalities in time-series data.

1: data processing system 10: storage device
20: display device 100: data processing device
110: time series acquisition unit 120: feature extraction unit
130: first division 140: second division
150: abnormality detection unit 190: output unit
401: processor 402: memory
403 processing circuit

Claims (7)

a time series acquisition unit that acquires time series data;
a feature extraction unit for extracting a feature amount of the time-series data based on the time-series data acquired by the time-series acquisition unit;
a first dividing unit dividing the time-series data into a plurality of first partial time-series data based on the feature amount extracted by the feature extracting unit;
a second dividing unit for dividing each of the plurality of first partial time-series data divided by the first dividing unit into a plurality of second partial time-series data based on the feature amount extracted by the feature extracting unit;
An anomaly detection unit configured to detect an abnormality in the time-series data acquired by the time-series acquisition unit based on at least one of the plurality of second partial time-series data divided by the second division unit,
Each of the plurality of first partial time-series data divided by the first dividing unit has a first oscillation interval time-series data and a first stable interval time-series data,
The second divider converts the first oscillation interval time-series data among the first partial time-series data into a plurality of second partial time-series data for each of the plurality of first partial time-series data divided by the first divider. divide,
Each of the plurality of second partial time-series data divided by the second dividing unit has second oscillation interval time-series data and second stable interval time-series data,
The anomaly detection unit detects anomaly in the time-series data acquired by the time-series acquisition unit based on the second stable interval time-series data among a plurality of second partial time-series data divided by the second division unit.
Characterized by a data processing device. According to claim 1,
The feature extraction unit extracts a first feature from the time-series data acquired by the time-series acquisition unit using a first slide window having a first predetermined time length;
The first division unit divides the time-series data into a plurality of first partial time-series data based on the first feature extracted by the feature extraction unit;
The feature extraction unit obtains a second feature from each of the plurality of pieces of first partial time-series data divided by the first dividing unit using a second slide window having a predetermined second time length shorter than the first time length. extract,
The second dividing unit divides the first partial time-series data into a plurality of second partial time-series data based on the extracted second feature.
Characterized by a data processing device. According to claim 1,
The anomaly detection unit determines that the second stable interval time-series data among the plurality of second partial time-series data divided by the second division unit and the second stable interval time-series data are at a number position in the first oscillation interval time-series data. Detecting anomalies in the time-series data acquired by the time-series acquisition unit based on second stable interval positional information indicating whether the second stable interval time-series data is the second stable interval time-series data of
Characterized by a data processing device. According to claim 3,
The anomaly detection unit, based on the second stable interval time-series data and the second stable interval position information, converts the second stable interval time-series data into the first oscillation of the first partial time-series data different from the second division unit. As the second stable interval time-series data among the plurality of second partial time-series data obtained by dividing the interval time-series data, the time-series acquisition unit compares the second stable interval time-series data with the same second stable interval position information. Detecting anomalies in the time series data to be acquired
Characterized by a data processing device.
delete delete delete

Images