Information processing apparatus, information processing method, and program

Abstract

To provide an information processing apparatus, an information processing method, and a program used to estimate a state of a device.SOLUTION: According to one embodiment, an information processing apparatus includes learning means configured to learn a local waveform pattern and a state estimator used to estimate a state of a device, based on multiple pieces of first sub-time series data divided from first time series data representing a waveform based on a base cycle of the waveform of a physical quantity changing in accordance with an operation of the device.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] Embodiments of the present invention relate to an information processing apparatus, an information processing method, and a program.

Background Art

[0002] Generally, in order to estimate the state of a device (high-frequency device) such as a bearing, a grinding machine, a robot arm, and a motor, frequency analysis is performed on time-series data (physical quantities representing waveforms in the time-series data) measured in the device.

[0003] However, although frequency analysis can capture spectra, it is difficult to detect slight changes in the shape of the waveform itself represented by the time-series data, and it is not possible to estimate (detect) the state of the device early from local shape changes in the waveform.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] Therefore, the problem to be solved by the present invention is to provide an information processing apparatus, an information processing method, and a program used for estimating the state of a device.

Means for Solving the Problems

[0006] The information processing apparatus according to the embodiment includes learning means for learning a local waveform pattern and a state estimator used for estimating the state of the device based on a plurality of first partial time-series data divided from first time-series data representing the waveform based on a base period of the waveform of a physical quantity that changes according to the operation of the device.

Brief Description of Drawings

[0007]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Figure 15

Figure 16

Figure 17

Figure 18

Figure 19

Figure 20

Embodiments for Carrying Out the Invention

[0008] Hereinafter, each embodiment will be described with reference to the drawings. (First Embodiment) First, the first embodiment will be described. The information processing apparatus according to the present embodiment is used to estimate the state of a device (hereinafter referred to as a target device) in which characteristics corresponding to a state such as normal or abnormal appear in frequency (that is, the state appears at a specific cycle). The target devices in the present embodiment are assumed to be high-frequency devices such as bearings, grinding machines, robot arms, and motors.

[0009] FIG. 1 is a block diagram showing an example of the functional configuration of the information processing apparatus according to the present embodiment. As shown in FIG. 1, the information processing apparatus 10 includes a first storage unit 11, a base period specifying unit 12, a time series data dividing unit 13, a learning unit 14, and a second storage unit 15.

[0010] The first storage unit 11 corresponds to a database that stores time series data representing the waveform of a physical quantity that changes according to the operation of the target device. It is assumed that a plurality of time series data are stored in the first storage unit 11.

[0011] The base period specifying unit 12 acquires the time series data stored in the first storage unit 11 and specifies the base period of the waveform represented by the time series data. The base period specified by the base period specifying unit 12 is output to the time series data dividing unit 13 and the learning unit 14.

[0012] The time-series data division unit 13 acquires the time-series data stored in the first storage unit 11, and divides the time-series data into a plurality of partial time-series data (division data) based on the base period output from the base period specifying unit 12. The plurality of partial time-series data divided from the time-series data by the time-series data division unit 13 are output to the learning unit 14.

[0013] The learning unit 14 performs learning of the local waveform pattern and the state estimator used for estimating the state of the target device based on the plurality of partial time-series data output from the time-series data division unit 13. Although detailed description is omitted, the length of the local waveform pattern is determined based on the base period output from the base period specifying unit 12.

[0014] The second storage unit 15 corresponds to a database that stores the learning results (that is, the local waveform pattern and the state estimator) by the learning unit 14.

[0015] FIG. 2 shows an example of the hardware configuration of the information processing apparatus 10. As shown in FIG. 2, the information processing apparatus 10 includes a CPU 101, a non-volatile memory 102, a main memory 103, an input device 104, a display device 105, a communication device 106, and the like.

[0016] The CPU 101 is a hardware processor that controls the operations of the components within the information processing apparatus 10. The CPU 101 may be composed of a single processor or a plurality of processors. The CPU 101 executes various programs loaded from the non-volatile memory 102, which is a storage device, to the main memory 103. The programs executed by the CPU 101 include an operating system (OS) and various application programs.

[0017] The input device 104 is a device configured to input various data, and includes, for example, a mouse, a keyboard, and the like. The display device 105 is a device configured to display various data, and includes, for example, a display and the like. The communication device 106 is a device configured to perform communication with an external device, for example, by wire or wirelessly.

[0018] In FIG. 2, only the non-volatile memory 102 and the main memory 103 are shown, but the information processing apparatus 10 may further include other storage devices such as an HDD (Hard Disk Drive) and an SSD (Solid State Drive).

[0019] In the present embodiment, the first storage unit 11 and the second storage unit 15 shown in FIG. 1 are realized by, for example, the non-volatile memory 102 shown in FIG. 2 or other storage devices.

[0020] Also, in the present embodiment, part or all of the base period specifying unit 12, the time series data dividing unit 13, and the learning unit 14 shown in FIG. 1 are realized by causing the CPU 101 shown in FIG. 2 to execute a predetermined program, that is, by software. This program may be downloaded to the information processing apparatus 10 via a network, or may be stored in a storage medium and distributed.

[0021] Here, although part or all of the units 12 to 14 have been described as being realized by software, part or all of the units 12 to 14 may be realized by hardware such as an IC (Integrated Circuit), or may be realized by a configuration combining software and hardware.

[0022] Next, the operation of the information processing apparatus 10 according to the present embodiment will be described. The information processing apparatus 10 according to the present embodiment operates as a time series data analysis apparatus that learns a local waveform pattern corresponding to a part (partial waveform) important for capturing a change in the state of a target device from waveforms represented by time series data stored in the first storage unit 11 and a state estimator for estimating the state of the target device based on the local waveform pattern.

[0023] First, with reference to FIG. 3, an example of the data structure (table in the database) of the time series data stored in the first storage unit 11 will be described.

[0024] In the example shown in FIG. 3, a plurality of time series data including time series data 111 and 112 are stored in the first storage unit 11, and each of the time series data is associated with an id and includes values of t1 to t T .

[0025] The id is identification information (assigned to the time series data) for identifying the time series data. t1 to t T corresponds to time, and the values of t1 to t T are physical quantities measured at times t1 to t T while the target device is operating. In other words, it can be said that in the first storage unit 11, the values of T times are recorded for each instance. Such time series data can represent a waveform by plotting the values of t1 to t T included in the time series data in association with the times t1 to t T .

[0026] Note that, for example, when the target device is a bearing, the values of t1 to t T are accelerations representing vibrations generated by the operation of the bearing, and are measured by, for example, an acceleration sensor attached to the bearing.

[0027] In this embodiment, it is assumed that the plurality of time-series data (physical quantities representing waveforms) stored in the first storage unit 11 are data measured when an object device in a normal state (for example, the object device before or immediately after the start of operation) is operated.

[0028] FIG. 4 is a flowchart showing an example of the processing procedure of the information processing apparatus 10 according to this embodiment.

[0029] First, the base period specifying unit 12 specifies the base period of the time-series data stored in the first storage unit 11 (step S1).

[0030] Hereinafter, with reference to the flowchart of FIG. 5, the processing of step S1 shown in FIG. 4 (processing of the base period specifying unit 12) will be described in detail.

[0031] First, the base period specifying unit 12 acquires one of the plurality of time-series data stored in the first storage unit 11 (step S11). Hereinafter, the time-series data acquired in step S11 is referred to as target time-series data.

[0032] Next, the base period specifying unit 12 calculates a power spectrum by applying a method such as the periodogram method or the Welch method to the target time-series data (step S12). In step S12, when performing discrete Fourier transform, if the values at the start point and the end point of the target time-series data do not match, in order to suppress the leakage error caused by the influence of the discontinuous points seemingly existing at the boundary and the estimated value of the spectrum spreading from the original peak, the power spectrum may be calculated while using a window function such as a Hann window or a Hanning window.

[0033] The power spectrum calculated in step S12 represents the power (intensity) for each frequency of the target time-series data, and the base period specifying unit 12 specifies the period corresponding to the frequency at which the power is maximum in the power spectrum (the reciprocal of the frequency) as the base period (step S13).

[0034] When the process of step S13 is executed, it is determined whether or not the processes of steps S11 to S13 described above have been executed for all the time-series data stored in the first storage unit 11 (step S14).

[0035] If it is determined that the process has not been executed for all the time-series data (NO in step S14), the process returns to step S11 and is repeated. In this case, the time-series data for which the processes of steps S11 to S13 have not been executed is acquired in step S11, and the processes of steps S12 and S13 are executed with the acquired time-series data as the target time-series data.

[0036] On the other hand, assume that it is determined that the process has been executed for all the time-series data (YES in step S14). In this case, by repeatedly executing the processes of steps S11 to S13 for each of the time-series data stored in the first storage unit 11, the base period is specified (acquired) for each time-series data. Therefore, the base period specifying unit 12 outputs a representative value of the base period specified for each time-series data in this way (step S15).

[0037] It is assumed that the representative value of the base period output in step S15 is a value calculated by, for example, weighted average (weighted mean), but it may be other values such as the median or the mode.

[0038] According to the process shown in FIG. 5 described above, even when a plurality of time-series data are stored (accumulated) in the first storage unit 11, one base period (a representative value of the base period specified for each time-series data) is output based on the plurality of time-series data.

[0039] In FIG. 5, the period corresponding to the frequency at which the power is maximum in the power spectrum was described as being specified as the base period. However, the base period may be specified as, for example, the period corresponding to the frequency calculated by weighted average for a plurality of frequencies corresponding to the top power in the power spectrum.

[0040] In other words, the base period in the present embodiment may be specified using the frequencies with the top power (that is, at least one frequency) in the power spectrum calculated based on the time-series data.

[0041] Also, although here it was described that a window function is used for calculating the power spectrum, a window function may not be used for calculating the power spectrum.

[0042] Furthermore, in FIG. 5, it was described that the power spectrum is calculated. However, the base period may be specified without using the power spectrum. Specifically, for example, the autocorrelation coefficient is calculated based on a plurality of plots (t1 to t T values) on the waveform represented by the time-series data, and the interval (temporal length) obtained from two plots (corresponding times) with the maximum autocorrelation coefficient among the plurality of plots may be specified as the base period. Note that the base period may be calculated by weighted average or the like for the intervals obtained for every two plots with the top autocorrelation coefficient.

[0043] In other words, the base period in the present embodiment may be specified using the plots with the top autocorrelation coefficient in the waveform represented by the time-series data.

[0044] Returning to FIG. 4 again, the time-series data division unit 13 divides the time-series data stored in the first storage unit 11 based on the base period output from the base period specification unit 12 by executing the process shown in FIG. 5 described above (step S2).

[0045] Hereinafter, with reference to the flowchart of FIG. 6, the process of step S2 (the process of the time series data division unit 13) shown in FIG. 4 will be described in detail.

[0046] First, the time series data division unit 13 acquires the base period output from the base period specification unit 12 (step S21).

[0047] Next, the time series data division unit 13 acquires one of the plurality of time series data stored in the first storage unit 11 (step S22). Hereinafter, the time series data acquired in step S22 is referred to as target time series data.

[0048] When the process of step S22 is executed, the time series data division unit 13 divides the target time series data based on the base period acquired in step S21 and a predetermined magnification (step S23).

[0049] Here, when the base period is B and the predetermined magnification is C, the time series data division unit 13 divides the target time series data into a plurality of partial time series data each having a data length of B×C. According to this, for example, as shown in FIG. 3, when the time series data includes values at T time points (that is, the number of measurements of the physical quantity in the time series data is T), by executing the process of step S23, the target time series data is divided into T / (B×C) rounded-down number of partial time series data (division data). In other words, each of the plurality of partial time series data corresponds to time series data including values at B×C time points.

[0050] Note that the above-mentioned predetermined magnification is assumed to be, for example, 10 times or the like, but other magnifications may also be used. Also, the predetermined magnification is assumed to be specified by a user who uses the information processing apparatus 10 via an input device 104 or the like, for example, but it may be preset inside the information processing apparatus 10 (time series data division unit 13).

[0051] When the process of step S23 is executed, it is determined whether the processes of steps S22 and S23 described above have been executed for all the time-series data stored in the first storage unit 11 (step S24).

[0052] If it is determined that the process has not been executed for all the time-series data (NO in step S24), the process returns to step S22 and is repeated. In this case, the time-series data for which the processes of steps S22 and S23 have not been executed is acquired in step S22, and the process of step S23 is executed with the time-series data as the target time-series data.

[0053] On the other hand, if it is determined that the process has been executed for all the time-series data (YES in step S24), the time-series data splitting unit 13 outputs a plurality of partial time-series data obtained by splitting each of the plurality of time-series data stored in the first storage unit 11 (that is, a plurality of partial time-series data created by executing the process of step S23 for each time-series data) (step S25).

[0054] Returning to FIG. 4 again, the learning unit 14 performs learning of the local waveform pattern and the state estimator based on the base period output from the base period specifying unit 12 by executing the process shown in FIG. 5 described above and the plurality of partial time-series data output from the time-series data splitting unit 13 by executing the process shown in FIG. 6 described above (step S3).

[0055] Hereinafter, with reference to the flowchart of FIG. 7, the process of step S3 (the process of the learning unit 14) shown in FIG. 4 will be described in detail.

[0056] First, the learning unit 14 acquires the base period output from the base period specifying unit 12 and the plurality of partial time-series data output from the time-series data splitting unit 13 (step S31).

[0057] Here, assume that, for example, Shapelets learning is used for the learning of the local waveform pattern and the state estimator in the present embodiment. In this case, the local waveform pattern corresponds to the pattern of representative partial waveforms included in the time series data and is referred to as Shapelets.

[0058] In the present embodiment, the length of the local waveform pattern is determined based on the base period acquired in step S31. As an example, the learning unit 14 sets the length of the local waveform pattern to the base period (step S32).

[0059] Next, the learning unit 14 performs the learning of the local waveform pattern and the state estimator using the plurality of partial time series data acquired in step S31 (step S33). The learning of the local waveform pattern and the state estimator in the present embodiment is performed by applying the OCLTS (One Class Learning Time-series Shapelets) related technology to the plurality of partial time series data. OCLTS is a time series waveform anomaly detection method with explanatory power that can be learned only from normal cases. Specifically, according to OCLTS, by learning the local waveform pattern (Shapelets) from the normal waveform, it is possible to realize detecting an anomaly when deviating from the local waveform (the waveform is broken compared to the local waveform).

[0060] Note that the learning of the local waveform pattern in step S33 means, for example, updating the shape of the local waveform pattern by fitting the local waveform pattern to the partial time series data (the waveform represented by the partial time series data). Since the length (time length) of the local waveform pattern is shorter than the length (time length) of the partial time series data, the learning unit 14 shifts the local waveform pattern in the time axis direction of the partial time series data, compares the waveform shapes of the local waveform pattern and the partial time series data, and changes the shape of the local waveform pattern according to the shape of the partial waveform of the partial time series data that is most similar to the local waveform pattern.

[0061] Also, assuming that the data measured when operating the target device in a normal state with the time-series data (physical quantity representing a waveform) stored in the first storage unit 11 as described above, the learning of the state estimator in step S33 means, for example, updating the parameters (e.g., weights, etc.) of the state estimator so as to output a score (hereinafter referred to as a state score) based on the degree of deviation between the partial time-series data (waveform represented thereby) and the local waveform pattern when the partial time-series data is input. Note that, as the state estimator in the present embodiment, for example, OCSVM (One Class Support Vector Machine) can be used.

[0062] Here, although the OCLTS related technology has been described as being applied for the local waveform pattern and the learning of the state estimator, for example, learning may be performed such that the local waveform pattern is the average value of the partial waveforms corresponding to one cycle of the partial time-series data, or learning of the state estimator may be performed to capture the change in the amplitude of the waveform of one cycle (that is, output a state score based on the change in the amplitude of the waveform of one cycle).

[0063] When the process of step S33 is executed, the learning unit 14 outputs, as the processing result of step S33, the learned local waveform pattern and the state estimator (parameters thereof) (step S34).

[0064] Returning to FIG. 4 again, the local waveform pattern and the state estimator output from the learning unit 14 are stored in the second storage unit 15 by executing the process shown in FIG. 7 described above (step S4).

[0065] Here, generally, state estimation of a device may be performed by performing frequency analysis on time-series data. Although such frequency analysis can capture the spectrum, this frequency analysis is not a technique for detecting a slight change in the shape of the waveform itself represented by the time-series data, and it is difficult to capture a change in which the amplitude increases only for several cycles at a relatively low frequency in a device that gradually deteriorates.

[0066] Specifically, for example, a case can be considered where the waveform represented by time-series data is disrupted by irregular collisions of the balls of a bearing at a relatively low frequency. However, in the frequency analysis described above, such a change in the waveform cannot be captured as a sign of abnormality in this case.

[0067] On the other hand, the information processing apparatus 10 according to the present embodiment performs learning of local waveform patterns and state estimators used for estimating the state of the target device based on a plurality of partial time-series data (first partial time-series data) divided from time-series data (first time-series data) representing the waveform based on the base period of the waveform of the physical quantity that changes according to the operation of the target device.

[0068] According to the local waveform patterns and state estimators learned in the present embodiment, it is considered possible to estimate the state of the target device corresponding to the sign of abnormality described above based on the time-series data measured when the target device in operation is operating.

[0069] Specifically, when the target device in the present embodiment is, for example, a bearing that gradually deteriorates, data acquired at the start of operation or the like for the bearing (that is, time-series data measured when a bearing in a normal state is operated) is used to perform learning of the local waveform pattern and the abnormality prediction detection model (state estimator). By doing so, it is possible to capture a change in the waveform shape (local change in the shape of the waveform) that rarely appears in several cycles as a sign of abnormality using the local waveform pattern and the abnormality prediction detection model, and to detect an early change in the state of the bearing.

[0070] Therefore, it can be said that the information processing apparatus 10 according to the present embodiment is useful for estimating the state of the device (detecting a sign of abnormality).

[0071] In this embodiment, as shown in FIG. 8, for example, by dividing the time-series data (waveform represented thereby) 200 into a plurality of partial time-series data (waveforms represented thereby) 201 to 204, it becomes possible to increase the number of data (number of samples) used for learning (that is, improve the learning efficiency). In this case, each of the plurality of partial time-series data 201 to 204 can have a length obtained by multiplying, for example, a multiple specified by the user with respect to the base period.

[0072] Also, in this embodiment, by performing learning (for example, Shapelets learning) based on the plurality of partial time-series data described above, the shape of the local waveform pattern and the parameters of the state estimator are updated, and the estimation accuracy of the state of the target device using the local waveform pattern and the state estimator can be improved.

[0073] Also, in this embodiment, the length of the local waveform pattern is determined based on the base period, but the base period may be specified based on the frequency with the highest power in the power spectrum calculated based on the time-series data, or may be specified based on the plot with the highest autocorrelation coefficient in the waveform represented by the time-series data.

[0074] In this embodiment, the information processing apparatus 10 has been described as including each of the units 11 to 15 shown in FIG. 1, but the information processing apparatus 10 may have a configuration in which some of the units 11 to 15 shown in FIG. 1 are omitted. Specifically, for example, at least one of the first storage unit 11 and the second storage unit 15 may be arranged outside the information processing apparatus 10.

[0075] Also, the information processing apparatus 10 according to this embodiment assumes a case where it is realized by one device, but may have a configuration in which the units 11 to 15 are respectively arranged in separate devices (that is, may be realized by a plurality of devices).

[0076] (Second Embodiment) Next, the second embodiment will be described. In this embodiment, the parts different from the above-described first embodiment will be mainly described.

[0077] In the above-described first embodiment, it was described that the local waveform pattern and the state estimator are learned (that is, the information processing apparatus has only the function of performing learning). However, in this embodiment, the state of the target device is estimated using the local waveform pattern and the state estimator for which the learning has been performed (that is, the information processing apparatus has the function of monitoring the target device), which is different from the first embodiment.

[0078] FIG. 9 is a block diagram showing an example of the functional configuration of the information processing apparatus according to this embodiment. In FIG. 9, the same parts as those in FIG. 1 described above are denoted by the same reference numerals, and the detailed description thereof is omitted.

[0079] As shown in FIG. 9, the information processing apparatus 10 further includes a state estimation unit 16 and a display processing unit 17 in addition to the respective units 11 to 15 shown in FIG. 1.

[0080] The state estimation unit 16 acquires time-series data, and estimates the state of the target device using the time-series data, the local waveform pattern stored in the second storage unit 15, and the state estimator.

[0081] The display processing unit 17 displays the state of the target device (that is, the estimation result) estimated by the state estimation unit 16.

[0082] Here, the functional configuration of the information processing apparatus 10 according to this embodiment has been described with reference to FIG. 9. However, since the hardware configuration of the information processing apparatus 10 is the same as that of the first embodiment described above, the detailed description thereof is omitted here.

[0083] Note that part or all of the state estimation unit 16 and the display processing unit 17 shown in FIG. 9 described above may be realized by causing the CPU 101 shown in FIG. 2 to execute a predetermined program, that is, by software, or by hardware, or by a configuration combining software and hardware.

[0084] Here, in addition to the processing shown in FIG. 4 described above (hereinafter referred to as learning processing), the information processing apparatus 10 according to the present embodiment executes processing for estimating the state of the target device (hereinafter referred to as state estimation processing).

[0085] Hereinafter, with reference to the flowchart of FIG. 10, an example of the processing procedure of the above-described state estimation processing will be described. Note that the state estimation processing is executed after the learning processing, and when the state estimation processing is executed, it is assumed that the local waveform pattern and the state estimator learned by executing the learning processing are stored in the second storage unit 15.

[0086] Assuming that the time-series data (the time-series data stored in the first storage unit 11) used in the learning processing shown in FIG. 4 described above is learning data, the state estimation unit 16 acquires time-series data different from the learning data (hereinafter referred to as state estimation data) (step S41). The state estimation data acquired in step S41 corresponds to, for example, time-series data representing the waveform of a physical quantity (for example, acceleration, etc.) measured while the target device in operation is operating (time-series data for monitoring the target device in operation), and has the same data structure as the learning data.

[0087] Next, the time-series data splitting unit 13 splits the state estimation data acquired in step S41 (step S42). Since the processing in step S42 is the same as the processing in step S2 shown in FIG. 4 described above, the detailed description thereof is omitted here. When the processing in step S42 is executed, the state estimation unit 16 acquires a plurality of partial time-series data output from the time-series data splitting unit 13.

[0088] Note that the base period used in the process of step S42 is the base period specified by the base period specifying unit 12 in the learning process, and it is assumed to be held inside the time-series data division unit 13 when the learning process is executed.

[0089] Also, although the process of step S42 has been described here as being executed by the time-series data division unit 13, the process of step S42 may be executed by the state estimation unit 16.

[0090] When the process of step S42 is executed, the state estimation unit 16 applies the local waveform pattern and the state estimator stored in the second storage unit 15 to the plurality of partial time-series data divided from the state estimation data as described above, thereby obtaining a state score for each of the partial time-series data (step S43).

[0091] In step S43, a state score output from the state estimator is obtained by inputting a plurality of partial time-series data and the local waveform pattern to the state estimator. In this case, in the state estimator, the local waveform pattern is applied to each of the plurality of partial time-series data, and a process of calculating a state score based on the degree of deviation between the partial time-series data (the waveform represented thereby) and the local waveform pattern is executed.

[0092] Note that assuming that the local waveform pattern and the state estimator are learned based on the time-series data (learning data) measured when the target device in the normal state is operated as described in the first embodiment above, the above-described state score becomes a low value when the target device is in the normal state, and becomes a high value when the target device is in an abnormal state (that is, deviates from the normal state).

[0093] Here, in step S43 described above, a state score is obtained for each partial time-series data. The state estimation unit 16 calculates a representative value of the state scores obtained for each partial time-series data (step S44). Note that the representative value calculated in step S44 includes statistical quantities such as the average value, standard deviation, or maximum value of the state scores for each partial time-series data.

[0094] Next, the state estimation unit 16 estimates the state of the target device based on the representative value of the state scores calculated in step S44 (step S45). In step S45, for example, when the representative value of the state scores is equal to or greater than a predetermined value (threshold value), it can be estimated that a change has occurred in the state of the target device compared to the time when the learning data was measured (that is, there is a sign of abnormality in the target device). That is, the state score based on the degree of deviation from the local waveform pattern described above can be called an abnormality prediction score.

[0095] Note that in step S45, the process of estimating the state of the target device based on the state scores obtained for each partial time-series data may be executed, and other processes than those described above may be executed.

[0096] When the process of step S45 is executed, the state estimation unit 16 outputs the state of the target device estimated in step S45 (hereinafter referred to as the estimation result) (step S46).

[0097] Note that the estimation result output from the state estimation unit 16 is displayed on the display device 105 by, for example, the display processing unit 17.

[0098] Here, FIG. 11 shows an example of a display screen (hereinafter referred to as an estimation result display screen) when the estimation result is displayed on the display device 105.

[0099] As shown in FIG. 11, on the estimation result display screen, as an estimation result, it is displayed that there is a sign of abnormality in the target device. Note that a state score (a representative value of the state score calculated in step S44 described above) may be displayed on the estimation result display screen.

[0100] Furthermore, as shown in FIG. 11, on the estimation result display screen, as the basis for estimating the state of the target device, for example, the partial time-series data (the waveform represented thereby) in which the highest state score was obtained and the local waveform pattern are superimposed and displayed. In the example shown in FIG. 11, the partial time-series data is indicated by a thin line, and the local waveform pattern is indicated by a thick line. The deviation (degree of deviation) between the partial time-series data and the local waveform pattern thus superimposed and displayed is grasped by the user as the basis for the above-described estimation result.

[0101] Note that the estimation result display screen shown in FIG. 11 is an example, and the estimation result may be displayed in a manner different from that in FIG. 11. Specifically, the estimation result, the state score, and the basis for estimation may be, for example, processed and displayed on the estimation result display screen. Also, part of the estimation result, the state score, and the basis for estimation displayed on the estimation result display screen may be omitted, or information other than the estimation result, the state score, and the basis for estimation may be displayed on the estimation result display screen.

[0102] Here, the description has been made assuming that the estimation result output from the state estimation unit 16 is displayed on the display device 105, but the estimation result may be transmitted to, for example, a server device or a terminal device outside the information processing device 10 via the communication device 106.

[0103] As described above, the information processing apparatus 10 according to the present embodiment divides the state estimation data (second time-series data) based on the base period into a plurality of partial time-series data (second partial time-series data), and inputs the plurality of partial time-series data and the local waveform pattern into the state estimator, thereby obtaining, for each of the partial time-series data, a state score (a state score based on the degree of deviation between the partial time-series data and the local waveform pattern) output from the state estimator, and estimating the state of the target device based on the obtained state score (for example, a representative value of the state scores obtained for each partial time-series data).

[0104] In the present embodiment, with the above-described configuration, it is possible to estimate the state of the target device in operation, for example, using the local waveform pattern and the state estimator for which the learning described in the first embodiment above has been performed.

[0105] Note that in the present embodiment, as described in the first embodiment above, by setting the length of the local waveform pattern to the base period, it is possible to improve the state estimation accuracy of the target device.

[0106] Specifically, FIG. 12 shows a case where the length of the local waveform pattern is set to be longer than the base period. By using the local waveform pattern, 0.5 is calculated as a representative value of the state score, and it is shown that the target device is estimated to be in a normal state based on the state score.

[0107] On the other hand, FIG. 13 shows a case where the length of the local waveform pattern is set to the base period. By using the local waveform pattern, 2.3 is calculated as a representative value of the state score, and it is shown that it is estimated that there is a sign of abnormality in the target device based on the state score.

[0108] That is, as shown in FIG. 12, when the length of the local waveform pattern is set to be longer than the base period, abnormal portions in the time-series data (partial time-series data) cannot be appropriately captured by the local waveform pattern (that is, the abnormal portions become blurred with respect to the local waveform pattern, and it is impossible to grasp the omen of abnormality).

[0109] On the other hand, as shown in FIG. 13, when the length of the local waveform pattern is set to the base period, abnormal portions in the time-series data can be pinpointed by the local waveform pattern, so that it becomes possible to improve the accuracy of detecting (estimating) the omen of abnormality.

[0110] Furthermore, in the present embodiment, by estimating the state of the target device from the state scores based on each of the plurality of partial time-series data divided from the state estimation data, the robustness of the state estimation (that is, the evaluation of the state estimation data) can be improved.

[0111] Also, in the present embodiment, when the state (estimation result) of the target device estimated as described above is displayed, the partial time-series data and the local waveform pattern are superimposed and displayed. According to such a configuration, the user can easily grasp the (deviation) between the superimposed partial time-series data and the local waveform pattern as the basis for the estimation result of the target device.

[0112] Note that, in the present embodiment, the information processing apparatus 10 has been described as executing both the learning process and the state estimation process. However, the information processing apparatus 10 may be configured to execute only the state estimation process (that is, it does not have a function of performing learning and has only a function of monitoring the target device).

[0113] (Third Embodiment) Next, the third embodiment will be described. In the present embodiment, mainly the differences from the above-described second embodiment will be described.

[0114] This embodiment is different from the above-described second embodiment in that time-series data is segmented for each frequency band in the learning process and the state estimation process.

[0115] FIG. 14 is a block diagram showing an example of the functional configuration of the information processing apparatus according to this embodiment. In FIG. 14, the same parts as those in FIG. 9 described above are denoted by the same reference numerals, and detailed description thereof is omitted.

[0116] As shown in FIG. 14, the information processing apparatus 10 further includes a segmentation unit 18 in addition to the respective units 11 to 17 shown in FIG. 9.

[0117] The segmentation unit 18 segments (decomposes) time-series data (learning data and state estimation data) for each frequency band. In this embodiment, the learning of the local waveform pattern and the state estimator is performed for each frequency band in which the learning data is segmented by the segmentation unit 18. Further, the state of the target device in this embodiment is estimated for each frequency band in which the state estimation data is segmented by the segmentation unit 18.

[0118] Here, the functional configuration of the information processing apparatus 10 according to this embodiment has been described with reference to FIG. 14. However, since the hardware configuration of the information processing apparatus 10 is the same as that of the second embodiment described above, detailed description thereof is omitted here.

[0119] Note that part or all of the segmentation unit 18 shown in FIG. 14 described above may be realized by causing the CPU 101 shown in FIG. 2 to execute a predetermined program, that is, by software, or by hardware, or by a configuration combining software and hardware.

[0120] Hereinafter, the learning process and the state estimation process executed by the information processing apparatus 10 according to this embodiment will be described.

[0121] FIG. 15 is a flowchart showing an example of the processing procedure of the learning process executed in the present embodiment.

[0122] First, the segmentation unit 18 segments the time-series data (learning data) stored in the first storage unit 11 (step S51).

[0123] Hereinafter, with reference to the flowchart of FIG. 16, the process of step S51 shown in FIG. 15 (the process of the segmentation unit 18) will be described in detail.

[0124] First, the segmentation unit 18 acquires, for example, the number of segments specified by the user (step S511). Although it has been described here as acquiring the number of segments specified by the user, in step S511, for example, the number of segments pre-held inside the information processing apparatus 10 (segmentation unit 18) may be acquired.

[0125] Next, the segmentation unit 18 acquires one of the plurality of learning data (step S512). Hereinafter, the learning data acquired in step S512 is referred to as target learning data.

[0126] Next, the segmentation unit 18 obtains the Nyquist frequency of the target learning data, and divides (equally divides) from 0 Hz to the Nyquist frequency by the number of segments acquired in step S511 (step S513). Note that the Nyquist frequency corresponds to the maximum frequency detectable by the Fast Fourier Transform.

[0127] The segmentation unit 18 performs filtering of the target learning data using a band-pass filter (step S514) in order to extract time-series data of one of a plurality of frequency bands (hereinafter referred to as the target frequency band) obtained by dividing from 0 Hz to the Nyquist frequency in step S513 from the target learning data. Note that the time-series data of the target frequency band obtained by performing filtering in step S514 is referred to as a segment.

[0128] When the process of step S514 is executed, it is determined whether or not the process of step S514 has been executed for all the frequency bands obtained by dividing from 0 Hz to the Nyquist frequency (step S515).

[0129] If it is determined that the process has not been executed for all the frequency bands (NO in step S515), the process returns to step S514 and is repeated. In this case, the process of step S514 is executed with the frequency band for which the process of step S514 has not been executed as the target frequency band.

[0130] On the other hand, if it is determined that the process has been executed for all the frequency bands (YES in step S515), it is determined whether or not the processes of steps S512 to S515 described above have been executed for all the learning data (step S516).

[0131] If it is determined that the process has not been executed for all the learning data (NO in step S516), the process returns to step S512 and is repeated. In this case, the learning data for which the processes of steps S512 to S515 have not been executed is acquired in step S512, and the processes of steps S513 to S515 are executed with the acquired learning data as the target learning data.

[0132] On the other hand, when it is determined that the processing has been executed for all the learning data (YES in step S516), the segmentation unit 18 outputs segments (time-series data for each frequency band) extracted from each of the learning data by repeatedly executing the processing in step S514 (step S517).

[0133] In the present embodiment, it has been described that the learning data is segmented for each frequency band (that is, time-series data for each frequency band is extracted from the learning data) by executing the processing described in FIG. 16 above. However, the method of segmentation may be different from the method described here.

[0134] Specifically, here it has been described that the frequency band corresponding to each segment is obtained by dividing from 0 Hz to the Nyquist frequency by the number of segments. However, the minimum and maximum values of the frequency may be adjusted as appropriate. Also, the frequency band corresponding to each segment may be divided by the number of segments in an arbitrary range specified by the user, for example, or may be directly specified by the user. Furthermore, the frequency bands corresponding to each segment may partially overlap.

[0135] Furthermore, here it has been described that each segment is extracted using a band-pass filter. However, the segment may be obtained by converting the learning data into the frequency domain using Fourier transform (fast Fourier transform), then extracting only the specified frequency band and performing inverse transformation. Also, the segment may be obtained by representing the learning data as wavelet coefficients by continuous wavelet transform or discrete wavelet transform, for example, and then reconstructing using only some decomposition levels. Note that discrete wavelet transform includes multiresolution analysis or wavelet packets, etc.

[0136] That is, the segmentation of the learning data in the present embodiment may be a process of extracting time-series data in a specific frequency band from the learning data, and may be performed using at least one of a band-pass filter, Fourier transform, wavelet transform (continuous wavelet transform or discrete wavelet transform), etc.

[0137] Returning to FIG. 15 again, for each of the above-described segments (corresponding frequency bands), the processes of the following steps S52 to S55 are executed. Note that, assuming the frequency band corresponding to the segment for which the processes of steps S52 to S55 are executed is the target frequency band, the processes of steps S52 to S55 correspond to the process of using the learning data (time-series data) in the processes of steps S1 to S4 shown in FIG. 4 described above as the time-series data (i.e., segment) of the target frequency band. Therefore, the detailed description thereof is omitted here.

[0138] When the process of step S55 is executed, it is determined whether the processes of steps S52 to S55 described above have been executed for all segments (step S56).

[0139] If it is determined that the processes have not been executed for all segments, the process returns to step S52 and the processes are repeated.

[0140] On the other hand, if it is determined that the processes have been executed for all segments (YES in step S56), the learning process is terminated.

[0141] According to the above-described learning process, local waveform patterns and state estimators are learned for each segment (corresponding frequency band).

[0142] FIG. 17 is a flowchart showing an example of the processing procedure of the state estimation process executed in the present embodiment.

[0143] First, the process of step S61 corresponding to the process of step S41 shown in FIG. 10 described above is executed.

[0144] Next, the segmentation unit 18 segments the state estimation data acquired in step S61 (step S62). Since the process of step S62 corresponds to the process of using the learning data as the state estimation data in the process of step S51 shown in FIG. 15 described above, the detailed description thereof is omitted here.

[0145] When the process of step S62 is executed, for each segment (corresponding frequency band) extracted from the state estimation data by the execution of the process of step S62, the processes of the following steps S63 to S66 are executed. When the frequency band corresponding to the segment for which the processes of steps S63 to S66 are executed is the target frequency band, the processes of steps S63 to S66 correspond to the process of using the state estimation data (time series data) in the processes of S42 to S45 shown in FIG. 10 described above as the time series data (that is, the segment) of the target frequency band, and thus the detailed description thereof is omitted here.

[0146] When the process of step S66 is executed, it is determined whether or not the processes of steps S63 to S66 described above have been executed for all segments (step S67).

[0147] If it is determined that the process has not been executed for all segments (NO in step S67), the process returns to step S63 and is repeated.

[0148] On the other hand, if it is determined that the process has been executed for all segments (YES in step S67), the state estimation unit 16 outputs the state of the target device estimated for each segment in step S66 described above (that is, the estimation result for each frequency band corresponding to the segment) (step S68).

[0149] Note that the estimation result for each frequency band corresponding to the segment output from the state estimation unit 16 is displayed on the display device 105 by, for example, the display processing unit 17.

[0150] Here, FIG. 18 shows an example of an estimated result display screen in this embodiment. As shown in FIG. 18, on the estimated result display screen, the estimated result, the state score, and the basis for estimation are displayed in association with each of the frequency bands corresponding to the segments.

[0151] In the example shown in FIG. 18, the estimated result displayed on the estimated result display screen in association with frequency band 1 is the state of the target device estimated based on the time-series data of frequency band 1 extracted from the state estimation data when the state estimation data is segmented. Also, in the example shown in FIG. 18, the state score displayed on the estimated result display screen in association with frequency band 1 is the representative value of the state scores calculated when the state of the target device is estimated based on the time-series data of frequency band 1 extracted from the state estimation data when the state estimation data is segmented. Further, in the example shown in FIG. 18, on the estimated result display screen in association with frequency band 1, as the basis for estimation, the partial time-series data (partial time-series data divided from the time-series data of frequency band 1) in which the highest state score was obtained when the state of the target device was estimated based on the time-series data of frequency band 1 extracted from the state estimation data when the state estimation data is segmented and the local waveform pattern learned based on the time-series data of frequency band 1 are superimposed and displayed.

[0152] Here, only the frequency band 1 (the segment corresponding thereto) has been described, but on the estimated result display screen shown in FIG. 18, the estimated result, the state score, and the basis for estimation are similarly displayed for frequency bands 2 and 3 other than the frequency band 1.

[0153] In the example of FIG. 18, the estimated result, the state score, and the basis for estimation are displayed on the estimated result display screen in the order of frequency bands 1 to 3. However, as shown in FIG. 19, on the estimated result display screen, the estimated results with larger state scores may be preferentially displayed.

[0154] As described above, in this embodiment, the learning data (first time-series data) is segmented for each frequency, the state estimation data (second time-series data) is segmented for each frequency, and the learning of the local waveform pattern and the state estimator is performed for each frequency band in which the learning data is segmented, and the state of the target device is estimated for each frequency band in which the state estimation data is segmented.

[0155] In this embodiment, with such a configuration, by restricting the frequency band of the time-series data, even when the time-series data contains various frequency bands (frequency components), the necessary frequency band can be extracted to estimate the state of the target device.

[0156] Specifically, the Shapelets learning that learns the local waveform pattern from the time-series data (normal data) measured when the target device in a normal state is operated and performs anomaly detection from the change in the shape of the waveform using the local waveform pattern is often applied to devices (waveforms) that operate relatively stably. However, in this embodiment, it can be easily applied to devices in which waveform changes appear only in some frequency bands.

[0157] In this embodiment, although it has been described that a plurality of segments are extracted from the time-series data by segmentation and the state of the target device is independently estimated based on each of the segments, at least one segment may be extracted from the time-series data.

[0158] Also, the segmentation of the time-series data in this embodiment can be realized by, for example, a process of applying a band-pass filter for each frequency specified by the user to the time-series data, a process of performing inverse transformation for some frequency bands after converting the time-series data into the frequency domain by Fourier transform, and a process of performing inverse transformation for some decomposition levels after converting the time-series data into wavelet coefficients by continuous wavelet transform or discrete wavelet transform.

[0159] Furthermore, in the present embodiment, for example, by displaying the states of the target device estimated for each frequency band in the order of state scores, the user can easily grasp the frequency band that represents the change in the state of the target device (that is, a sign of abnormality appears).

[0160] In the present embodiment, since the frequency band in which a sign of abnormality of the target device appears can be grasped (specified), for example, when estimating the state of a device of the same type as the target device, only the time-series data of the grasped frequency band may be used (extracted).

[0161] (Fourth Embodiment) Next, the fourth embodiment will be described. In the present embodiment, mainly the parts different from the above-described first embodiment will be described.

[0162] In the above-described first embodiment, the learning process was described as being executed using the time-series data (physical quantity representing a waveform) measured when the target device in the normal state was operated. However, this embodiment is different from the first embodiment in that the learning process is executed using the time-series data measured when the target device in a plurality of states is operated.

[0163] Since the functional configuration and hardware configuration of the information processing apparatus according to the present embodiment are the same as those of the above-described first embodiment, the detailed description thereof will be omitted here, and in the following, it will be described with reference to FIG. 1 and the like as appropriate.

[0164] FIG. 20 shows an example of the data structure of the time-series data (learning data) stored in the first storage unit 11 in the present embodiment.

[0165] In the example shown in FIG. 20, a plurality of time-series data including time-series data 113 to 116 are stored in the first storage unit 11, and each of the time-series data includes a state label and t1 to t associated with an id. T The values of. Note that id and t1 to t TSince the value of is as described in the first embodiment (FIG. 3) described above, a detailed description thereof will be omitted here.

[0166] Here, in the present embodiment, the time-series data (t1 to t T of the value) is measured when the target device is operated, and the state label represents the state (class) of the target device when the time-series data is measured. The state label includes, for example, "0" representing a normal state, "1" representing an abnormal state A, "2" representing an abnormal state B, and the like. Note that abnormal state B is assumed to be a different type of abnormality from abnormal state A.

[0167] The learning process in the present embodiment is generally as described in FIG. 4, but learning (supervised learning) of the local waveform pattern and the state estimator is performed based on the state label included in the time-series data described above.

[0168] Specifically, in step S3 shown in FIG. 4, for each state label, learning is performed to update the shape of the local waveform pattern by fitting the local waveform pattern to the partial time-series data (that is, the partial time-series data divided from the time-series data including the state label). Also, the local waveform pattern and the state estimator may be simultaneously learned so that the classification performance is improved using all the state labels.

[0169] Also, in step S3 shown in FIG. 4, for each state label, learning is performed to update the parameters of the state estimator so as to output a state score based on the degree of deviation between the partial time-series data (that is, the waveform represented by the partial time-series data) and the local waveform pattern when the partial time-series data is input.

[0170] The state estimator trained in this embodiment can output a state score for each state of the target device when, for example, partial time-series data divided from state-estimation data is input. In this embodiment, based on the state scores for each state output from the state estimator in this way, a plurality of states of the target device (for example, normal, sign of abnormality A, sign of abnormality B, etc.) can be estimated.

[0171] Note that, in this embodiment, techniques such as Learning Time-series Shapelets and Region Of Interest may be applied.

[0172] Also, in this embodiment, it has been mainly described that the time-series data stored in the first storage unit 11 in the first embodiment described above includes state labels (that is, applicable to the first embodiment), but this embodiment may be applied to the second or third embodiment. Note that the state labels are included in the learning data for learning and are not included in the state-estimation data.

[0173] According to at least one of the embodiments described above, an information processing apparatus, an information processing method, and a program used for estimating the state of a device can be provided.

[0174] Although several embodiments of the present invention have been described, these embodiments are presented as examples and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the gist of the invention. These embodiments and their modifications are included in the scope and gist of the invention, and are also included in the invention described in the claims and its equivalent scope.

[0175] Regarding the above-described embodiments, the following additional remarks are disclosed. [1] An information processing apparatus including learning means for learning a local waveform pattern and a state estimator used for estimating the state of the device based on a plurality of first partial time series data divided from first time series data representing the waveform based on the base period of the waveform of a physical quantity that changes according to the operation of the device. [2] The learning of the local waveform pattern and the state estimator includes updating the shape of the local waveform pattern and the parameters of the state estimator. The information processing apparatus according to [1]. [3] The learning of the local waveform pattern and the state estimator is performed using Shapelets learning. The information processing apparatus according to [1] or [2]. [4] The length of the local waveform pattern is determined based on the base period. The information processing apparatus according to any one of [1] to [3]. [5] The base period is specified based on the frequency with the highest power in the power spectrum calculated based on the first time series data. The information processing apparatus according to any one of [1] to [4]. [6] The base period is specified based on the plot with the highest autocorrelation coefficient among a plurality of plots on the waveform represented by the first time series data. The information processing apparatus according to any one of [1] to [4]. [7] Each of the plurality of first partial time series data has a length obtained by multiplying the base period by a multiple specified by the user. The information processing apparatus according to any one of [1] to [6]. [8] Dividing means for dividing second time series data different from the first time series data into a plurality of second partial time series data based on the base period; Obtaining means for obtaining, for each of the second partial time series data, a score based on the degree of deviation between the second partial time series data output from the state estimator and the local waveform pattern by inputting the plurality of second partial time series data and the local waveform pattern into the state estimator; Estimation means for estimating the state of the device based on the obtained score The information processing apparatus according to any one of [1] to [7], further comprising [9] The estimation means estimates the state of the device based on a representative value of the scores obtained for each of the second partial time series data. The information processing apparatus according to [8].

[10] The information processing apparatus according to any one of [8] to [9], further comprising display processing means for displaying the estimated state of the device The display processing means superimposes and displays the second partial time series data and the local waveform pattern The information processing apparatus according to [8] or [9].

[11] The information processing apparatus according to any one of [8] to

[10] , further comprising segmentation means for segmenting the first time series data for each frequency band and segmenting the second time series data for each frequency band The learning of the local waveform pattern and the state estimator is performed for each frequency band into which the first time series data is segmented The state of the device is estimated for each frequency band into which the second time series data is segmented The information processing apparatus according to [8] to

[10] .

[12] The information processing apparatus according to

[11] , wherein the segmentation of the first and second time series data includes one of a process of applying a band-pass filter for each frequency band specified by a user to the first and second time series data, a process of performing an inverse transform for some frequency bands after converting the first and second time series data into a frequency domain by Fourier transform, and a process of performing an inverse transform for some decomposition levels after converting the first and second time series data into wavelet coefficients by continuous wavelet transform or discrete wavelet transform

[13] The information processing apparatus according to

[11] or

[12] , further comprising display processing means for displaying the state of the device estimated for each frequency band in the order of the scores

[14] The first time-series data includes a label representing the state of the device when a physical quantity representing a waveform in the first time-series data is measured. The learning means learns the state estimator so as to output a score capable of estimating a plurality of states of the device based on the label included in the first time-series data. The information processing apparatus according to any one of [1] to

[13] .

[15] An information processing method for estimating the state of a device, comprising: learning a local waveform pattern and a state estimator based on a plurality of first partial time-series data divided from first time-series data representing a waveform based on a base period of the waveform of a physical quantity that changes according to the operation of the device.

[16] The learning of the local waveform pattern and the state estimator includes updating the shape of the local waveform pattern and the parameters of the state estimator. The information processing method according to

[15] .

[17] The learning of the local waveform pattern and the state estimator is performed using Shapelets learning. The information processing method according to

[15] or

[16] .

[18] The length of the local waveform pattern is determined based on the base period. The information processing method according to any one of

[15] to

[17] .

[19] The base period is specified based on a frequency with a higher power in a power spectrum calculated based on the first time-series data. The information processing method according to any one of

[15] to

[18] .

[20] The base period is specified based on plots with higher autocorrelation coefficients among a plurality of plots on the waveform represented by the first time-series data. The information processing method according to any one of

[15] to

[18] .

[21] Each of the plurality of first partial time-series data has a length obtained by multiplying a multiple specified by a user with respect to the base period. The information processing method according to any one of

[15] to

[20] .

[22] Based on the base period, dividing second time-series data different from the first time-series data into a plurality of second partial time-series data; Inputting the plurality of second partial time-series data and the local waveform pattern into the state estimator, and obtaining, for each of the second partial time-series data, a score based on a degree of deviation between the second partial time-series data output from the state estimator and the local waveform pattern; estimating the state of the device based on the obtained scores; The information processing method according to any one of

[15] to

[21] , further comprising the above.

[23] The estimating includes estimating the state of the device based on a representative value of the scores obtained for each of the second partial time-series data, according to the information processing method described in

[22] .

[24] further comprising displaying the estimated state of the device; The displaying includes displaying the second partial time-series data and the local waveform pattern in a superimposed manner, according to the information processing method described in

[22] or

[23] .

[25] segmenting the first time-series data for each frequency band; segmenting the second time-series data for each frequency band; further comprising the above; The learning between the local waveform pattern and the state estimator is performed for each frequency band into which the first time-series data is segmented; The state of the device is estimated for each frequency band into which the second time-series data is segmented. The information processing method according to

[22] to

[24] .

[26] ​The segmentation of the first and second time series data includes one of the following processes: applying a band-pass filter for each frequency band specified by the user to the first and second time series data; performing an inverse transformation on some frequency bands after converting the first and second time series data into the frequency domain by Fourier transformation; and performing an inverse transformation on some decomposition levels after converting the first and second time series data into wavelet coefficients by continuous wavelet transformation or discrete wavelet transformation. The information processing method described in

[25] .

[27] The information processing method described in

[25] or

[26] , further comprising displaying the states of the device estimated for each frequency band in the order of the scores.

[28] The first time series data includes a label representing the state of the device when a physical quantity representing a waveform in the first time series data is measured. Performing the learning includes training a state estimator to output scores capable of estimating a plurality of states of the device based on the labels included in the first time series data. The information processing method according to any one of

[15] to

[27] .

[29] A program for causing a computer to function as a learning means for learning a local waveform pattern and a state estimator used for estimating the state of a device based on a plurality of first partial time series data divided from first time series data representing a waveform based on a base period of the waveform of a physical quantity that changes according to the operation of the device.

[30] The learning of the local waveform pattern and the state estimator includes updating the shape of the local waveform pattern and the parameters of the state estimator. The program described in

[29] .

[31] The learning of the local waveform pattern and the state estimator is performed using Shapelets learning. The program described in

[29] or

[30] .

[32] The length of the local waveform pattern is determined based on the base period, and is a program described in any one of

[29] to

[31] .

[33] The base period is specified based on the frequency with the highest power in the power spectrum calculated based on the first time series data, and is a program described in any one of

[29] to

[32] .

[34] The base period is specified based on the plot with the highest autocorrelation coefficient among a plurality of plots on the waveform represented by the first time series data, and is a program described in any one of

[29] to

[32] .

[35] Each of the plurality of first partial time series data has a length obtained by multiplying a multiple specified by the user with respect to the base period, and is a program described in any one of

[29] to

[34] .

[36] The computer is a dividing means for dividing the second time series data different from the first time series data into a plurality of second partial time series data based on the base period; an acquisition means for acquiring, for each of the second partial time series data, a score based on the degree of deviation between the second partial time series data output from the state estimator by inputting the plurality of second partial time series data and the local waveform pattern into the state estimator; and a program described in any one of

[29] to

[35] for further causing the computer to function as an estimation means for estimating the state of the device based on the acquired score.

[37] The estimation means estimates the state of the device based on a representative value of the scores acquired for each of the second partial time series data, and is a program described in

[36] .

[38] The computer is further caused to function as a display processing means for displaying the estimated state of the device, and the display processing means superimposes and displays the second partial time series data and the local waveform pattern. A program described in

[36] or

[37] .

[39] Further cause the computer to function as segmentation means for segmenting the first time-series data for each frequency band and segmenting the second time-series data for each frequency band, The learning of the local waveform pattern and the state estimator is performed for each frequency band in which the first time-series data is segmented, The state of the device is estimated for each frequency band in which the second time-series data is segmented

[36] ~

[38] The program according to any one of the above items.

[40] The segmentation of the first and second time-series data includes a process of applying a band-pass filter for each frequency band specified by the user to the first and second time-series data, a process of performing an inverse transform on some frequency bands after converting the first and second time-series data into the frequency domain by Fourier transform, and a process of performing an inverse transform on some decomposition levels after converting the first and second time-series data into wavelet coefficients by continuous wavelet transform or discrete wavelet transform. The program according to

[39] .

[41] Further cause the computer to function as display processing means for displaying the state of the device estimated for each frequency band in the order of the scores. The program according to

[39] or

[40] .

[42] The first time-series data includes a label representing the state of the device when a physical quantity representing a waveform in the first time-series data is measured, The learning means performs learning of the state estimator so as to output scores capable of estimating a plurality of states of the device based on the labels included in the first time-series data

[29] ~

[41] The program according to any one of the above items.

Explanation of symbols

[0176] 10... Information processing apparatus, 11... First storage unit, 12... Base period specifying unit, 13... Time-series data dividing unit, 14... Learning unit, 15... Second storage unit, 16... State estimation unit, 17... Display processing unit, 18... Segmentation unit, 101... CPU, 102... Non-volatile memory, 103... Main memory, 104... Input device, 105... Display device, 106... Communication device.

Claims

1. An information processing apparatus including a learning unit that learns a local waveform pattern and a state estimator used to estimate a state of a device based on a plurality of first partial time series data divided from first time series data representing a waveform based on a base period of the waveform of a physical quantity that changes according to the operation of the device.

2. The information processing apparatus according to claim 1, wherein learning of the local waveform pattern and the state estimator includes updating a shape of the local waveform pattern and parameters of the state estimator.

3. The information processing apparatus according to claim 1, wherein learning of the local waveform pattern and the state estimator is performed using Shapelets learning.

4. The information processing apparatus according to claim 1, wherein a length of the local waveform pattern is determined based on the base period.

5. The information processing apparatus according to claim 1, wherein the base period is specified based on a frequency with a higher power in a power spectrum calculated based on the first time series data.

6. The information processing apparatus according to claim 1, wherein the base period is specified based on plots with higher autocorrelation coefficients among a plurality of plots on a waveform represented by the first time series data.

7. The information processing apparatus according to claim 1, wherein each of the plurality of first partial time series data has a length obtained by multiplying a multiple specified by a user with respect to the base period.

8. A dividing unit that divides second time series data different from the first time series data into a plurality of second partial time series data based on the base period; An acquisition unit that acquires, for each of the second partial time series data, a score based on a degree of deviation between the second partial time series data output from the state estimator by inputting the plurality of second partial time series data and the local waveform pattern into the state estimator; An estimation unit that estimates a state of the device based on the acquired score The information processing apparatus according to claim 1, further including.

9. The information processing apparatus according to claim 8, wherein the estimation unit estimates the state of the device based on a representative value of the scores acquired for each of the second partial time series data.

10. The information processing apparatus further includes a display processing unit that displays the estimated state of the device, The display processing unit superimposes and displays the second partial time series data and the local waveform pattern The information processing apparatus according to claim 8.

11. Segmentation means for segmenting the first time series data for each frequency band and segmenting the second time series data for each frequency band is further provided. The learning of the local waveform pattern and the state estimator is performed for each frequency band in which the first time series data is segmented. The state of the device is estimated for each frequency band in which the second time series data is segmented. The information processing apparatus according to claim 8.

12. The segmentation of the first and second time series data includes one of a process of applying a band-pass filter for each frequency band specified by a user to the first and second time series data, a process of performing an inverse transform for some frequency bands after converting the first and second time series data into a frequency domain by Fourier transform, and a process of performing an inverse transform for some decomposition levels after converting the first and second time series data into wavelet coefficients by continuous wavelet transform or discrete wavelet transform. The information processing apparatus according to claim 11.

13. The information processing apparatus according to claim 11, further comprising display processing means for displaying the state of the device estimated for each frequency band in the order of the scores.

14. The first time series data includes a label representing the state of the device when a physical quantity representing a waveform in the first time series data is measured. The learning means performs learning of the state estimator so as to output scores capable of estimating a plurality of states of the device based on the labels included in the first time series data. The information processing apparatus according to claim 1.

15. An information processing method including performing learning of a local waveform pattern and a state estimator used for estimating the state of a device based on a plurality of first partial time series data divided from first time series data representing a waveform based on a base period of the waveform of a physical quantity that changes according to the operation of the device.

16. The learning of the local waveform pattern and the state estimator includes updating the shape of the local waveform pattern and the parameters of the state estimator. The information processing method according to claim 15.

17. The learning of the local waveform pattern and the state estimator is performed using Shapelets learning. The information processing method according to claim 15.

18. The length of the local waveform pattern is determined based on the base period. The information processing method according to claim 15.

19. The information processing method according to claim 15, wherein the base period is specified based on a frequency having a top power in a power spectrum calculated based on the first time series data.

20. The information processing method according to claim 15, wherein the base period is specified based on plots having top autocorrelation coefficients among a plurality of plots on a waveform represented by the first time series data.

21. The information processing method according to claim 15, wherein each of the plurality of first partial time series data has a length obtained by multiplying a multiple specified by a user with respect to the base period.

22. Based on the base period, dividing second time series data different from the first time series data into a plurality of second partial time series data; Obtaining, for each of the second partial time series data, a score based on a degree of deviation between the second partial time series data output from the state estimator and the local waveform pattern by inputting the plurality of second partial time series data and the local waveform pattern into the state estimator; Estimating the state of the device based on the obtained score The information processing method according to claim 15, further comprising.

23. The information processing method according to claim 22, wherein the estimating includes estimating the state of the device based on a representative value of the scores obtained for each of the second partial time series data.

24. Further comprising displaying the estimated state of the device, The displaying includes displaying the second partial time series data and the local waveform pattern in a superimposed manner The information processing method according to claim 22.

25. Segmenting the first time series data for each frequency band; Segmenting the second time series data for each frequency band Further comprising, Learning between the local waveform pattern and the state estimator is performed for each frequency band into which the first time series data is segmented, The state of the device is estimated for each frequency band into which the second time series data is segmented The information processing method according to claim 22.

26. The segmentation of the first and second time series data includes one of the following processes: applying a band-pass filter for each frequency band specified by the user to the first and second time series data; performing an inverse transformation for some frequency bands after converting the first and second time series data into the frequency domain by Fourier transform; and performing an inverse transformation for some decomposition levels after converting the first and second time series data into wavelet coefficients by continuous wavelet transform or discrete wavelet transform. The information processing method according to claim 25.

27. The information processing method according to claim 25, further comprising displaying the state of the device estimated for each frequency band in the order of the scores.

28. The first time series data includes a label representing the state of the device when a physical quantity representing a waveform in the first time series data is measured. Performing the learning includes training a state estimator to output scores capable of estimating a plurality of states of the device based on the labels included in the first time series data. The information processing method according to claim 15.

29. A program for causing a computer to function as learning means for training a local waveform pattern and a state estimator used for estimating the state of a device based on a plurality of first partial time series data divided from first time series data representing a waveform based on a base period of the waveform of a physical quantity that changes according to the operation of the device.

30. The training of the local waveform pattern and the state estimator includes updating the shape of the local waveform pattern and the parameters of the state estimator. The program according to claim 29.

31. The training of the local waveform pattern and the state estimator is performed using Shapelets learning. The program according to claim 29.

32. The length of the local waveform pattern is determined based on the base period. The program according to claim 29.

33. The base period is specified based on the frequencies with the highest power in the power spectrum calculated based on the first time series data. The program according to claim 29.

34. The base period is specified based on the plots with the highest autocorrelation coefficients among a plurality of plots on the waveform represented by the first time series data. The program according to claim 29.

35. The program according to claim 29, wherein each of the plurality of first partial time series data has a length obtained by multiplying a multiple specified by a user with respect to the base period.

36. The computer is further caused to a dividing means for dividing, based on the base period, second time series data different from the first time series data into a plurality of second partial time series data; an obtaining means for obtaining, for each of the second partial time series data, a score based on a degree of deviation between the second partial time series data output from the state estimator by inputting the plurality of second partial time series data and the local waveform pattern into the state estimator; The program according to claim 29, for further causing the computer to function as an estimating means for estimating the state of the device based on the obtained score.

37. The program according to claim 36, wherein the estimating means estimates the state of the device based on a representative value of the scores obtained for each of the second partial time series data.

38. The computer is further caused to function as a display processing means for displaying the estimated state of the device, and the display processing means displays the second partial time series data and the local waveform pattern in a superimposed manner The program according to claim 36.

39. The computer is further caused to function as a segmentation means for segmenting the first time series data for each frequency band and segmenting the second time series data for each frequency band, learning between the local waveform pattern and the state estimator is performed for each frequency band in which the first time series data is segmented, and the state of the device is estimated for each frequency band in which the second time series data is segmented The program according to claim 36.

40. The segmentation of the first and second time series data includes one of a process of applying a band-pass filter for each frequency band specified by a user to the first and second time series data, a process of performing an inverse transform for some frequency bands after converting the first and second time series data into a frequency domain by a Fourier transform, and a process of performing an inverse transform for some decomposition levels after converting the first and second time series data into wavelet coefficients by a continuous wavelet transform or a discrete wavelet transform. The program according to claim 39.

41. The program according to claim 39, further causing the computer to function as display processing means for displaying the state of the device estimated for each frequency band in the order of the scores.

42. The first time series data includes a label representing the state of the device when a physical quantity representing a waveform is measured in the first time series data. The learning means performs learning of the state estimator so as to output scores capable of estimating a plurality of states of the device based on the labels included in the first time series data. The program according to claim 29.

Images