Data analysis apparatus, data analysis method, and program

Abstract

This invention accurately identifies abnormal sound from time-series data having various characteristics. A determination unit (11) determines a characteristic of time-series data. A selection unit (12) selects a method for analyzing the time-series data on the basis of the characteristic of the time-series data. An identification unit (13) identifies an abnormal sound included in the time-series data by using the selected method to analyze the time-series data.

Description

Technical Field

[0001] The present invention relates to a data analysis apparatus, a data analysis method, and a recording medium, and more particularly to a data analysis apparatus, a data analysis method, and a recording medium for analyzing time series data.

Background Art

[0002] In railway vehicles, automobile engine rooms, factories, etc., technologies for monitoring the operating states of devices and components have been studied. For example, in the related technology described in Patent Document 1, a learning model generated by machine learning is used to diagnose the operating state of a machine in real time based on physical quantities detected by sensors.

[0003] Furthermore, in an example of another related technology, non-negative matrix factorization (NMF) is used to analyze time series data. Specifically, in the related technology, time series data is converted into an amplitude spectrogram, and the spectrogram is decomposed into a basis matrix and an activation matrix. Then, by using the activation matrix as an acoustic feature quantity, abnormal sounds included in the time series data are identified.

Prior Art Documents

Patent Documents

[0004] Japanese Patent Application Laid-Open No. 2020-204937

Summary of the Invention

Problems to be Solved by the Invention

[0005] NMF approximates a non-negative matrix, which is a representation of a spectrogram, by the product of non-negative matrices of lower dimensions. Therefore, when the period of peaks in time series data is unstable, or in a noisy environment, etc., the related technology using NMF has a reduced accuracy in identifying abnormal sounds.

[0006] The present invention has been made in view of the above problems, and an object thereof is to accurately identify abnormal sounds from time-series data of various properties.

Means for Solving the Problems

[0007] A data analysis apparatus according to an aspect of the present invention includes a determination unit that determines the property of time-series data, a selection unit that selects a method for analyzing the time-series data based on the property of the time-series data, and an identification unit that identifies an abnormal sound included in the time-series data by analyzing the time-series data using the selected method.

[0008] In a data analysis method according to an aspect of the present invention, the property of time-series data is determined, a method for analyzing the time-series data is selected based on the property of the time-series data, and the time-series data is analyzed using the selected method to identify an abnormal sound included in the time-series data.

[0009] A recording medium according to an aspect of the present invention stores a program for causing a computer to determine the property of time-series data, select a method for analyzing the time-series data based on the property of the time-series data, and identify an abnormal sound included in the time-series data by analyzing the time-series data using the selected method.

Effects of the Invention

[0010] According to an aspect of the present invention, an abnormal sound can be accurately identified from time-series data of various properties.

Brief Description of the Drawings

[0011]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Embodiments for Carrying Out the Invention

[0012] Some embodiments for carrying out the present invention will be described below.

[0013] 〔Embodiment 1〕 Embodiment 1 will be described with reference to FIGS. 1 to 2.

[0014] (Abnormal Sound Identification Device 10) FIG. 1 is a block diagram showing the configuration of the abnormal sound identification device 10 according to the present Embodiment 1. As shown in FIG. 1, the abnormal sound identification device 10 includes a determination unit 11, a selection unit 12, an identification unit 13, and a provision unit 14.

[0015] The determination unit 11 determines the nature of the time series data. The determination unit 11 is an example of determination means.

[0016] In the first example, the determination unit 11 tracks the peaks of the time-series data. Here, the determination unit 11 can use a well-known peak tracking technique. The determination unit 11 measures the time from the first peak of the time-series data to the next peak. Subsequently, the determination unit 11 calculates the time from the second peak to the third peak. Repeatedly, the determination unit 11 calculates the time width (referred to as the period) between adjacent peaks of the time-series data. Thereafter, the determination unit 11 calculates the fluctuation of the period of the peaks in the time-series data. For example, the determination unit 11 calculates the variance or standard deviation of the period of the peaks in the time-series data as an index indicating the magnitude of the fluctuation of the period of the peaks in the time-series data. And when the magnitude of the fluctuation of the period of the peaks in the time-series data is equal to or less than the threshold value, the determination unit 11 determines that the time-series data has property a. On the other hand, when the magnitude of the fluctuation of the period of the peaks in the time-series data exceeds the threshold value, the determination unit 11 determines that the time-series data has property b (Embodiment 1).

[0017] In the second example, the determination unit 11 performs a Fourier transform on the time-series data into a spectrum. The determination unit 11 calculates the peak intensity of the spectrum. And when all the peak intensities of the spectrum are equal to or greater than the threshold value, the determination unit 11 determines that the time-series data has property a. On the other hand, when one or more peak intensities of the time-series data are below the threshold value, the determination unit 11 determines that the time-series data has property b (Embodiment 2). Note that the method by which the determination unit 11 determines the property of the time-series data is not limited to the first and second examples described here.

[0018] The determination unit 11 outputs the determination result of the property of the time-series data to the selection unit 12. Also, the determination unit 11 outputs the time-series data to the identification unit 13.

[0019] The selection unit 12 selects a method for analyzing the time-series data based on the property of the time-series data. The selection unit 12 is an example of selection means.

[0020] In one example, the selection unit 12 receives from the determination unit 11 the determination result of the nature of the time-series data. Based on the determination result of the nature of the time-series data, the selection unit 12 selects a method for analyzing the time-series data. For example, when the time-series data has property a, the selection unit 12 selects a first method using Nonnegative Matrix Factorization (NMF). In the method using Nonnegative Matrix Factorization (hereinafter referred to as NMF), a spectrogram in which the spectra of the time-series data are arranged in time order is decomposed into a basis matrix and an activation matrix. The activation matrix thus obtained is the feature quantity in the first method.

[0021] On the other hand, when the time-series data has property b, the selection unit 12 selects a second method using Mel-Frequency Cepstrum Coefficients (MFCC). Mel-Frequency Cepstrum Coefficients (hereinafter referred to as MFCC) are the weighted cepstrum low-order components obtained by performing cepstrum analysis on the time-series data. The MFCC thus obtained is the feature quantity in the second method. The selection unit 12 notifies the discrimination unit 13 of the method (the first method or the second method) for analyzing the time-series data.

[0022] The discrimination unit 13 discriminates abnormal sounds included in the time-series data by analyzing the time-series data using the selected method. The discrimination unit 13 is an example of a discrimination means.

[0023] In one example, the identification unit 13 receives time-series data from the determination unit 11. Further, the identification unit 13 is notified by the selection unit 12 of a method (either the first method or the second method) for analyzing the time-series data. The identification unit 13 analyzes the time-series data using the method selected by the selection unit 12. For example, when the first method is selected, the identification unit 13 first converts the time-series data into a spectrogram. Then, the identification unit 13 obtains an activation matrix by decomposing the spectrogram using NMF. The identification unit 13 inputs the obtained activation matrix as a feature quantity to a classifier (hereinafter referred to as classifier A) that has been machine-learned using the activation matrix as a feature quantity. Classifier A identifies the time-series data based on the feature quantity of the input activation matrix and outputs the identification result.

[0024] On the other hand, when the second method is selected, the identification unit 13 first obtains MFCCs by performing cepstrum analysis on the time-series data. The identification unit 13 inputs the MFCCs obtained by cepstrum analysis as a feature quantity to a classifier (hereinafter referred to as classifier B) that has been machine-learned using the MFCCs as a feature quantity. Classifier B identifies the time-series data based on the feature quantity of the input MFCCs and outputs the identification result. In this way, the identification unit 13 identifies the time-series data using classifier A or classifier B according to the method selected by the selection unit 12. The identification unit 13 may output the identification result of the time-series data to a subsequent processing unit (not shown) or provide it to a recording medium or an external device.

[0025] (Operation of the abnormal sound identification device 10) With reference to FIG. 2, the operation of the abnormal sound identification device 10 according to Embodiment 1 will be described. FIG. 2 is a flowchart showing the flow of processing executed by each part of the abnormal sound identification device 10.

[0026] First, time-series data is input into the abnormal sound identification device 10. The time-series data is, for example, an acoustic signal generated by collecting the sound emitted by a device or component with a microphone in a running railway vehicle, a factory, an engine room of an automobile, etc. The abnormal sound identification device 10 receives the time-series data such as the acoustic signal via an arbitrary network, either wireless or wired. Then, the abnormal sound identification device 10 starts the following operations.

[0027] As shown in FIG. 2, the determination unit 11 determines the nature of the time-series data (S1). In one example, the determination unit 11 measures the time width (period) from the peak of the time-series data to the next peak. And when the magnitude of the fluctuation of the period is equal to or less than the threshold value, the determination unit 11 determines that the time-series data has property a. On the other hand, when the magnitude of the fluctuation of the period exceeds the threshold value, the determination unit 11 determines that the time-series data has property b. The determination unit 11 outputs the determination result of the nature of the time-series data to the selection unit 12. Also, the determination unit 11 outputs the time-series data to the identification unit 13.

[0028] Next, the selection unit 12 selects a method for analyzing the time-series data based on the nature of the time-series data (S2). For example, when the time-series data has property a, the selection unit 12 selects the first method using NMF. Thereafter, when the time-series data has property b, the selection unit 12 selects the second method using MFCC. The selection unit 12 notifies the identification unit 13 of the method for analyzing the time-series data.

[0029] The identification unit 13 identifies the abnormal sound included in the time-series data by analyzing the time-series data using the method selected by the selection unit 12 (S3).

[0030] Thus, the operation of the abnormal sound identification device 10 according to the first embodiment ends.

[0031] (Effect of this embodiment) According to the configuration of this embodiment, the determination unit 11 determines the nature of the time-series data. The selection unit 12 selects a method for analyzing the time-series data based on the nature of the time-series data. The identification unit 13 identifies abnormal sounds included in the time-series data by analyzing the time-series data using the selected method. The time-series data includes various types of sounds (including abnormal sounds) and noises, and the nature of the time-series data also varies. For example, there may be abnormal sounds with large periodic fluctuations in the time-series data, or there may be cases where the noise is large (the target sound is small).

[0032] The abnormal sound identification device 10 first determines the nature of the time-series data, and based on the determination result, selects a method for analyzing the time-series data. Thereby, abnormal sounds can be accurately identified from time-series data of various natures.

[0033] 〔Embodiment 2〕 Referring to FIGS. 3 to 5, Embodiment 2 will be described. In this Embodiment 2, an example of a method for determining the nature of time-series data will be described. In this Embodiment 2, regarding the configuration described in Embodiment 1, the description of Embodiment 1 is cited and the description thereof is omitted.

[0034] (Abnormal sound identification device 20) FIG. 3 is a block diagram showing the configuration of the abnormal sound identification device 20 according to this Embodiment 2. As shown in FIG. 3, the abnormal sound identification device 20 includes a determination unit 21, a selection unit 12, and an identification unit 13. Further, the determination unit 21 of the abnormal sound identification device 20 includes a peak detection unit 24. The peak detection unit 24 detects the peaks of the time-series data.

[0035] Referring to FIG. 4, an example of a method for determining the nature of time-series data will be specifically described. FIG. 4 exemplifies time-series data having property a and time-series data having property b, respectively. In FIG. 4, the peaks detected by the peak detection unit 24 are indicated by dots (black-filled circles).

[0036] In the second embodiment, the determination unit 21 determines the nature of the time-series data based on the time width (referred to as the period) from when a peak of the time-series data is detected until the next peak is detected.

[0037] In FIG. 4, the period of the peak in the time-series data is represented by the distance between the points indicating the peaks of the time-series data (that is, the length of both arrows). In the upper time-series data, the period of the peak in the time-series data is almost constant. In other words, the upper time-series data has a small difference in period (fluctuation in period). On the other hand, in the lower time-series data, there is variation in the period of the peak in the time-series data. In other words, the lower time-series data has a large difference in period (fluctuation in period).

[0038] The determination unit 21 compares the magnitude of the difference in the period of the peak (fluctuation in period) in the time-series data with a predetermined threshold value. For example, when the magnitude of the fluctuation in the period of the peak in the time-series data is represented by the deviation of the difference, the threshold value X is 0.5. In this example, when the fluctuation in the period of the peak in the time-series data is X = 0.5 or less, the determination unit 21 determines that the time-series data has property a. On the other hand, when the fluctuation in the period of the peak in the time-series data exceeds X = 0.5, the determination unit 21 determines that the time-series data has property b.

[0039] (Operation of the abnormal sound discrimination device 20: S1) Referring to FIG. 5, the operation of the abnormal sound discrimination device 20 according to the second embodiment will be described. Here, only the details of the processing flow executed by the determination unit 21, that is, the content of step S1 shown in FIG. 2, will be described.

[0040] Similar to the first embodiment, the abnormal sound discrimination device 20 receives time-series data. Thereafter, the determination unit 21 of the abnormal sound discrimination device 20 determines the nature of the time-series data as described below.

[0041] As shown in FIG. 5, the peak detection unit 24 of the determination unit 21 detects the peak of the time-series data (S21).

[0042] Based on the time width between peaks of the time series data, the determination unit 21 calculates the fluctuation of the peak period in the time series data (S22).

[0043] The determination unit 21 determines whether the magnitude of the fluctuation of the peak period in the time series data is less than or equal to a threshold value (S23).

[0044] When the magnitude of the fluctuation of the peak period in the time series data is less than or equal to the threshold value (Yes in S23), the determination unit 21 determines that the time series data has property a (S24A). On the other hand, when the magnitude of the fluctuation of the peak period in the time series data exceeds the threshold value (No in S23), the determination unit 21 determines that the time series data has property b (S24B).

[0045] Thus, the processing of the determination unit 21 is completed. Thereafter, the process proceeds to the processing of the selection unit 12 (step S2) described in the first embodiment. In the second embodiment, the description after the processing of the selection unit 12 (step S2) is omitted.

[0046] (Effect of this embodiment) According to the configuration of this embodiment, the determination unit 21 determines the property of the time series data. The selection unit 12 selects a method for analyzing the time series data based on the property of the time series data. The identification unit 13 identifies the abnormal sound included in the time series data by analyzing the time series data using the selected method. The time series data includes various types of sounds (including abnormal sounds) and noises, and the properties of the time series data are also various. For example, there may be cases where the time series data includes abnormal sounds with large fluctuations in the period, or cases where the noise is large (the target sound is small).

[0047] The abnormal sound identification device 10 first determines the property of the time series data, and selects a method for analyzing the time series data based on the determination result. Thereby, the abnormal sound can be accurately identified from time series data of various properties.

[0048] Furthermore, according to the configuration of the second embodiment, the determination unit 21 includes a peak detection unit 24 that detects peaks in time-series data. The determination unit 21 determines the nature of the time-series data based on the time width from when a peak in the time-series data is detected until the next peak is detected. From the time width from when a peak in the time-series data is detected until the next peak is detected, it is possible to calculate the magnitude of the fluctuation of the peak period in the time-series data. Then, by comparing the magnitude of the fluctuation of the peak period in the time-series data with a threshold value, it is possible to determine the property that the fluctuation of the peak period in the time-series data is relatively small and the property that the fluctuation of the peak period in the time-series data is relatively large.

[0049] 〔Embodiment 3〕 Referring to FIGS. 6 to 9, Embodiment 3 will be described. In this Embodiment 3, another example of a method for determining the nature of time-series data will be described. In this Embodiment 3, regarding the configuration described in the first embodiment, the description of the first embodiment is cited and the description thereof is omitted.

[0050] (Abnormal sound discrimination device 30) FIG. 6 is a block diagram showing the configuration of the abnormal sound discrimination device 30 according to the third embodiment. As shown in FIG. 6, the abnormal sound discrimination device 30 includes a determination unit 31, a selection unit 12, and an identification unit 13. Further, the determination unit 31 of the abnormal sound discrimination device 30 includes a data conversion unit 34. The data conversion unit 34 converts a signal in the time domain such as time-series data or a waveform into a signal in the frequency domain such as a spectrum or a spectrogram. Hereinafter, an example of converting time-series data into a spectrogram will be described.

[0051] FIG. 7 shows an example of a spectrogram converted from time-series data. In the spectrogram shown in FIG. 7, the frequency spectrum intensity is represented by shading. Also, the peaks of the frequency spectrum are indicated by thick lines (bars). In the example shown in FIG. 7, some of the peaks of the frequency spectrum are inclined with respect to the vertical axis and the horizontal axis. This represents that the peak frequency is changing with time. In other words, the period of the peak (= 1 / peak frequency) in the original time-series data is fluctuating.

[0052] FIG. 8 is a graph showing an example of a frequency spectrum converted from time-series data. The frequency spectrum corresponds to a predetermined time width in the spectrogram. In FIG. 8, the peaks of the frequency spectrum are indicated by points (black-filled circles) on the graph.

[0053] The fact that the peak of the frequency spectrum is sharp and high corresponds to the fact that the period of the peak in the original time-series data is almost constant (i.e., the fluctuation of the period is small) within a predetermined time width. On the other hand, the fact that the peak of the frequency spectrum is dull and low corresponds to the fact that there is variation in the period of the peak in the original time-series data (i.e., the fluctuation of the period is large) within a predetermined time width.

[0054] The determination unit 31 determines the nature of the time-series data based on the peak intensity of the frequency spectrum cut out from the spectrogram for each predetermined time width. For example, the determination unit 31 calculates the difference between the peak intensity in the frequency spectrum and the average of the intensities in a predetermined band centered on the peak frequency. The determination unit 31 compares the obtained difference with a threshold value Y. In this example, when the difference between the peak intensity in the frequency spectrum and the average of the intensities in a predetermined band centered on the peak frequency is equal to or greater than the threshold value Y, the determination unit 31 determines that the time-series data has property a. On the other hand, when the difference between the peak intensity in the frequency spectrum and the average of the intensities in a predetermined time width is less than the threshold value Y, the determination unit 31 determines that the time-series data has property b.

[0055] Note that by feeding back the information on the reliability of the abnormal sound identification result by the subsequent identification unit 13 to the determination unit 31, the determination unit 31 may update the threshold value Y so that the reliability of the abnormal sound identification result by the identification unit 13 increases.

[0056] (Operation of the abnormal sound identification device 30: S1) Referring to FIG. 9, the operation of the abnormal sound identification device 30 according to the third embodiment will be described. Here, only the details of the processing flow executed by the determination unit 31, that is, the content of step S1 shown in FIG. 2, will be described.

[0057] Similar to the first embodiment, the abnormal sound identification device 30 receives time-series data. Thereafter, the determination unit 31 of the abnormal sound identification device 30 determines the nature of the time-series data as described below.

[0058] As shown in FIG. 9, the data conversion unit 34 of the determination unit 31 converts time-series data (a signal in the time domain) into a spectrogram (FIG. 7) (a signal in the frequency domain) (S31).

[0059] The determination unit 31 generates a frequency spectrum each time by cutting out a segment of a predetermined time width from the spectrogram. The determination unit 31 calculates the peak intensity in the frequency spectrum (S32).

[0060] The determination unit 31 determines whether or not the peak intensity in the frequency spectrum is equal to or greater than a threshold value (S33). For example, the threshold value is the average of the intensities in a predetermined band centered on the peak frequency.

[0061] When the peak intensity in the frequency spectrum is equal to or greater than the threshold value (Yes in S33), the determination unit 31 determines that the time-series data has property a (S34A). On the other hand, when the peak intensity in the frequency spectrum is less than the threshold value (No in S33), the determination unit 31 determines that the time-series data has property b (S34B).

[0062] With the above, the processing of the determination unit 31 ends. Thereafter, the process proceeds to the process of the selection unit 12 (step S2) described in the first embodiment. In the third embodiment, the description after the process of the selection unit 12 (step S2) is omitted.

[0063] (Method for Determining Threshold Value) Here, a configuration in which the determination unit 31 determines the nature of the time-series data by comparing the peak intensity of the frequency spectrum with the threshold value has been described. Here, an example of a method for determining the threshold value of the peak intensity of the frequency spectrum will be described.

[0064] FIG. 10 is an example of a graph showing the distribution of scores used to determine the threshold value of the peak intensity of the frequency spectrum.

[0065] The determination unit 31 calculates scores for a number of learning data that includes the same number or approximately the same number of time-series data determined to have property a and time-series data determined to have property b. The score here is the difference between the average intensity in a predetermined band centered on the peak frequency and the peak intensity. From the calculation results of the scores, a score distribution as shown in FIG. 10 is obtained. Then, the determination unit 31 determines a threshold value based on the score distribution so that the time-series data having property a and the time-series data having property b can be distinguished. For example, the determination unit 31 determines twice the maximum value of the scores of the time-series data having property b as the threshold value. When the score of a certain time-series data is equal to or higher than the threshold value, the probability that the time-series data has property a is high, while when the score of a certain time-series data is lower than the threshold value, the probability that the time-series data has property b is high. The determination unit 31 can determine the nature of the time-series data as described above by using the threshold value determined in this way.

[0066] (Effects of this Embodiment) According to the configuration of the present embodiment, the determination unit 31 determines the nature of the time-series data. The selection unit 12 selects a method for analyzing the time-series data based on the nature of the time-series data. The identification unit 13 identifies abnormal sounds included in the time-series data by analyzing the time-series data using the selected method. The time-series data includes various types of sounds (including abnormal sounds) and noises, and the nature of the time-series data also varies. For example, there may be a case where the time-series data includes an abnormal sound with a large fluctuation in period, or a case where the noise is large (the target sound is small).

[0067] The abnormal sound identification device 10 first determines the nature of the time-series data, and based on the determination result, selects a method for analyzing the time-series data. Thereby, abnormal sounds can be accurately identified from time-series data of various natures.

[0068] Furthermore, according to the configuration of the third embodiment, the determination unit 31 includes a data conversion unit 34 that converts the time-series data into a spectrogram. The determination unit 31 determines the nature of the time-series data based on the peak intensity of the frequency spectrum cut out every predetermined time width from the spectrogram. The sharpness and strength of the peak of the frequency spectrum correspond to the small fluctuation of the period of the peak in the original time-series data. Then, by comparing the peak intensity of the frequency spectrum with a threshold value, it is possible to determine the nature in which the fluctuation of the period of the peak in the time-series data is relatively small and the nature in which the fluctuation of the period of the peak in the time-series data is relatively large.

[0069] (Modification example) In a modification example of any one of the first to third embodiments, the identification unit 13 identifies abnormal sounds included in the time-series data using three or more identifiers.

[0070] For example, the identification unit 13 according to this modification example uses an MFCC as a feature quantity and learns by machine learning. Together with the identifier B, an identifier that uses DCTC (Discrete Cosine Transform Coefficients) as a feature quantity (hereinafter referred to as identifier C) is used in combination. The identification unit 13 identifies abnormal sounds included in the time series data by each of the two identifiers, and compares the magnitudes of the reliability of the identification results. When the reliability of the identification result by the identifier B is higher, the identification unit 13 outputs the identification result by the identifier B. On the other hand, when the reliability of the identification result by the identifier C is higher, the identification unit 13 outputs the identification result by the identifier C.

[0071] In another modification example, the identification unit 13 may selectively use a plurality of identifiers according to the location where the acoustic signal that is the source of the time series data is acquired. In this modification example, each identifier and information indicating mutually different locations are associated in advance. The identification unit 13 according to this modification example also receives, from the determination unit 11, information indicating the location associated with the time series data together with the time series data. The identification unit 13 selects the corresponding identifier from among the plurality of identifiers based on the information indicating the location. Then, the identification unit 13 uses the selected identifier to identify abnormal sounds included in the time series data.

[0072] According to the configuration of this modification example, since one of the two identifiers with a higher reliability of the identification result is selected, the reliability of the identification result output by the identification unit 13 can be improved.

[0073] 〔Hardware Configuration〕 Each component of the abnormal sound identification devices 10, 20, and 30 described in the first to third embodiments indicates a block of a functional unit. Some or all of these components are realized by an information processing device 900 as shown in FIG. 11, for example. FIG. 11 is a block diagram showing an example of the hardware configuration of the information processing device 900.

[0074] As shown in FIG. 11, the information processing device 900 includes, as an example, the following configuration.

[0075] ·CPU (Central Processing Unit) 901 ·ROM (Read Only Memory) 902 ·RAM (Random Access Memory) 903 · Program 904 loaded into RAM 903 · Storage device 905 for storing program 904 · Drive device 907 for reading and writing to recording medium 906 · Communication interface 908 connected to communication network 909 · Input / output interface 910 for inputting and outputting data · Bus 911 connecting each component Each component of the abnormal sound identification devices 10, 20, and 30 described in the above Embodiments 1 to 3 is realized by the CPU 901 reading and executing a program 904 that realizes these functions. The program 904 that realizes the functions of each component is, for example, stored in advance in the storage device 905 or the ROM 902, and is loaded into the RAM 903 and executed by the CPU 901 as necessary. Note that the program 904 may be supplied to the CPU 901 via the communication network 909, or may be stored in advance in the recording medium 906, and the drive device 907 may read the program and supply it to the CPU 901.

[0076] According to the above configuration, the abnormal sound identification devices 10, 20, and 30 described in the above Embodiments 1 to 3 are realized as hardware. Therefore, effects similar to those described in the above Embodiments 1 to 3 can be achieved.

[0077] 〔Supplementary Note〕 One aspect of the present invention is also described as follows in the supplementary note, but is not limited thereto.

[0078] (Supplementary Note 1) Determination means for determining the nature of time-series data, Selection means for selecting a method for analyzing the time-series data based on the nature of the time-series data, Discrimination means for discriminating abnormal sounds included in the time-series data by analyzing the time-series data using the selected method, and a data analysis device.

[0079] (Appendix 2) The determination means includes peak detection means for detecting peaks in the time-series data, and the determination means determines the property of the time-series data based on the time width from when the peak in the time-series data is detected until the next peak is detected. The data analysis device according to Appendix 1, characterized in that.

[0080] (Appendix 3) The determination means includes data conversion means for converting the time-series data into a spectrogram, and the determination means determines the property of the time-series data based on the peak intensity of the frequency spectrum cut out every predetermined time width from the spectrogram. The data analysis device according to Appendix 1, characterized in that.

[0081] (Appendix 4) The determination means determines the magnitude of the variation in the period of the periodic component included in the time-series data, and the selection means selects a method according to the magnitude of the variation in the period from among a plurality of methods for analyzing the time-series data. The data analysis device according to any one of Appendices 1 to 3, characterized in that.

[0082] (Appendix 5) When the magnitude of the variation in the period is less than a threshold value, the selection means selects NMF (Nonnegative Matrix Factorization). The data analysis device according to Appendix 4, characterized in that.

[0083] (Appendix 6) Determine the property of the time-series data, Based on the property of the time-series data, select a method for analyzing the time-series data, By analyzing the time-series data using the selected method, identify abnormal sounds included in the time-series data Data analysis method.

[0084] (Appendix 7) Determine the property of the time-series data, Based on the property of the time-series data, select a method for analyzing the time-series data, By analyzing the time-series data using the selected method, identify abnormal sounds included in the time-series data A non-transitory recording medium storing a program for causing a computer to execute the above.

Industrial Applicability

[0085] The present invention can be used, for example, in a noise discrimination device that discriminates abnormal sounds emitted from railways, engine rooms of automobiles, factories, and other equipment or parts.

Explanation of Signs

[0086] 10 Noise discrimination device 11 Judgment unit 12 Selection unit 13 Identification unit 20 Noise discrimination device 24 Peak detection unit 30 Noise discrimination device 34 Data conversion unit

Claims

1. Determination means for determining the nature of time-series data, Selection means for selecting a method for analyzing the time-series data based on the nature of the time-series data, Discrimination means for discriminating abnormal sounds included in the time-series data by analyzing the time-series data using the selected method, The determination means determines the magnitude of the variation in the period of the periodic component included in the time-series data, The selection means selects a method corresponding to the magnitude of the variation in the period from among a plurality of methods for analyzing the time-series data Data analysis device.

2. The determination means includes peak detection means for detecting peaks in the time-series data, The determination means determines the nature of the time-series data based on the time width from when a peak in the time-series data is detected until the next peak is detected. The data analysis device according to claim 1, characterized in that.

3. The determination means includes data conversion means for converting the time-series data into a spectrogram, The determination means determines the nature of the time-series data based on the peak intensity of the frequency spectrum cut out from the spectrogram for each predetermined time width. The data analysis device according to claim 1, characterized in that.

4. When the magnitude of the variation in the period is less than a threshold value, the selection means selects NMF (Nonnegative Matrix Factorization). The data analysis device according to any one of claims 1 to 3, characterized in that.

5. A computer Determines the nature of time-series data, Based on the nature of the time-series data, selects a method for analyzing the time-series data, A data analysis method for discriminating abnormal sounds included in the time-series data by analyzing the time-series data using the selected method, The computer Determines the magnitude of the variation in the period of the periodic component included in the time-series data, Selects a method corresponding to the magnitude of the variation in the period from among a plurality of methods for analyzing the time-series data Data analysis method.

6. Determining the nature of time-series data, Based on the nature of the time-series data, selecting a method for analyzing the time-series data, By analyzing the time-series data using the selected method, discriminating abnormal sounds included in the time-series data, A program for causing a computer to execute, determining the magnitude of the variation in the period of the periodic component included in the time series data, selecting, from among a plurality of methods for analyzing the time series data, a method according to the magnitude of the variation in the period, A program for causing the computer to execute.

Images