JP6880992B2 - Evaluation method, evaluation device and evaluation program - Google Patents

Description

The present invention relates to a sound data processing technique.

The vibration method is known as a technique for diagnosing the state of equipment such as production equipment and infrastructure equipment. However, in order to implement the vibration method, it is necessary to attach a vibration sensor to each equipment, and since this vibration sensor is expensive, it is not widely used due to the problem of cost balance between investment and recovery.

Acoustic diagnosis is known as a technique for solving this problem. Acoustic diagnosis is a technology for detecting abnormal noise in equipment by acoustic analysis of sound data. For example, in a certain acoustic diagnosis, the operation mode of the equipment is associated with each cluster of the feature amount extracted from the sound data by spectral decomposition, and the feature amount extracted from the target sound data is compared with each cluster. Therefore, it is determined whether or not an abnormal noise has occurred.

Therefore, abnormal noise can be detected if the operation mode is appropriately associated with each cluster of features, but the association may not be properly executed depending on the conventional technique for acoustic diagnosis.

Japanese Unexamined Patent Publication No. 2009-236645 Japanese Unexamined Patent Publication No. 2001-117580 Japanese Unexamined Patent Publication No. 2016-42117

Hisae Shibuya, Shunji Maeda, "Anomaly Detection Technology Using Operation Pattern Information", IEEJ Transactions on Electronics, Information and Systems (Journal of Electronics, Information and Systems), Institute of Electrical Engineers of Japan, October 1, 2013, No. 133 , Volume 10, pp. 1998-2006

An object of the present invention is, in one aspect, to provide a technique for correctly associating an operation mode with each cluster of features extracted from sound data.

In the evaluation method according to one aspect, the feature amount at each time is extracted from the sound data, a plurality of feature amount clusters are generated by clustering the extracted feature amount, and the feature amount is based on the change of the cluster to which the feature amount belongs at each time. The probability of transition from each of the multiple clusters to another cluster is calculated, and based on the probability of transition from each of the multiple clusters to another cluster, the target is in steady operation for each of the multiple clusters. The process includes associating the first data indicating that the target is present, the second data indicating that the target is in non-constant operation, or the third data indicating that the target is stopped.

On one aspect, it becomes possible to correctly associate the operation mode with each cluster of the feature amount extracted from the sound data.

FIG. 1 is a diagram showing an example of three clusters of features extracted from the measured sound data. FIG. 2 is a diagram showing the correspondence of the operation modes for each cluster. FIG. 3 is a diagram showing an example of three clusters of features extracted from the measured sound data. FIG. 4 is a diagram showing the correspondence of the operation modes for each cluster. FIG. 5 is a diagram showing an outline of the system of the present embodiment. FIG. 6 is a functional block diagram of the analyzer. FIG. 7 is a diagram showing a processing flow of processing for associating an operation mode with each cluster. FIG. 8 is a diagram showing an example of sound data. FIG. 9 is a diagram showing time-series data of feature quantities. FIG. 10 is a diagram showing an example of data stored in the cluster data storage unit. FIG. 11 is a diagram showing an example of the distribution of points corresponding to the feature amount. FIG. 12 is a diagram showing an example of the result of clustering. FIG. 13 is a diagram showing an example of data stored in the cluster data storage unit. FIG. 14 is a diagram for explaining the transition of the cluster. FIG. 15 is a diagram showing an example of data stored in the probability data storage unit. FIG. 16 is a diagram showing an example of data stored in the probability data storage unit. FIG. 17 is a diagram showing a processing flow of processing for associating an operation mode with each cluster. FIG. 18 is a diagram showing the transition of clusters. FIG. 19 is a diagram showing an example of data stored in the correspondence data storage unit. FIG. 20 is a diagram for explaining an effective value. FIG. 21 is a diagram showing an example of data stored in the correspondence data storage unit. FIG. 22 is a diagram showing three clusters of features when there is no environmental sound. FIG. 23 is a diagram showing three clusters of features when there is an environmental sound. FIG. 24 is a diagram showing a true cluster. FIG. 25 is a diagram showing an example of data stored in the probability data storage unit. FIG. 26 is a diagram showing the result of association by the method of the present embodiment. FIG. 27 is a diagram showing a processing flow of processing for determining whether or not abnormal noise is emitted from the equipment. FIG. 28 is a diagram for explaining the degree of abnormality. FIG. 29 is a diagram showing an example of data stored in the evaluation result storage unit. FIG. 30 is a functional block diagram of the computer.

The operation mode in the present embodiment is divided into a steady operation mode, a non-steady operation mode, and a stop mode. The sound measured in the steady operation mode includes a sound emitted when a rotating body of equipment (for example, a motor, an engine, etc.) or a device such as a compressor is operating stably. In the stop mode, since the equipment is stopped, no sound is emitted from the equipment, and only surrounding environmental sounds (for example, weather sounds, work sounds, neighborhood living sounds, etc.) are included. The unsteady mode is a mode through which the transition from the stop mode to the steady operation mode and the transition from the steady operation mode to the stop mode are performed. For example, the state during the start processing of the equipment or the stop processing of the equipment is in progress. It is a state. Therefore, the sound measured in the unsteady mode includes the sound emitted when the equipment of the equipment is not operating stably.

FIG. 1 is a diagram showing an example of three clusters of features extracted from the measured sound data. In the example of FIG. 1, the horizontal axis and the vertical axis represent the values of the feature quantities in each dimension, and clustering for the two-dimensional feature vectors is executed.

As described above, the "feature amount" to be clustered is usually represented not by a one-dimensional value but by a vector having two or more dimensions, but for the sake of simplicity, it is simply referred to as a "feature amount" below.

As one method of associating the operation mode with each cluster, a method of associating the operation mode based on the size of the cluster can be considered. Considering the feature that the cluster radius becomes large because the spectrum distribution fluctuates in the unsteady operation mode, in the case of the example of FIG. 1, the cluster 1 is associated with the unsteady operation mode. Further, considering the feature that the effective value of the sound wave type in the stop mode is smaller than the effective value of the sound wave type in the steady operation mode because there is no sound source in the stop mode, in the example of FIG. 1, the cluster 2 is in the stop mode. And the cluster 3 is associated with the steady operation mode. That is, as shown in FIG. 2, the operation modes are associated. Here, it is assumed that the true operation mode can be associated.

However, noise elements due to surrounding environmental sounds are mixed in the sound data, and the cluster radius and cluster center change due to the influence. FIG. 3 is a diagram showing another example of three clusters of features extracted from the measured sound data. Compared with the example of FIG. 1, the example of FIG. 3 changes the cluster radius and the cluster center due to the mixing of noise elements. In this case, the cluster 3 having the largest cluster radius is associated with the unsteady operation mode. Further, of the cluster 1 and the cluster 2, the cluster 2 having a larger effective value of the sound wave type is assigned to the steady operation mode, and the cluster 1 is assigned to the stop mode. That is, as shown in FIG. 4, the operation modes are associated. Here, since the true operation mode could not be associated with each other, it becomes impossible to properly detect the abnormal noise.

In this way, if the operation modes are associated using the cluster radius, an error in the association may occur. Therefore, a method of correctly associating the operation modes will be described below.

FIG. 5 is a diagram showing an outline of the system of the present embodiment. For example, the equipment 5 such as a production facility or an infrastructure facility is equipped with a sound data acquisition device 3 for acquiring sound data of a sound emitted from the facility 5 and an environmental sound. The sound data acquisition device 3 is connected to the analysis device 1 via a network, and transmits the acquired sound data to the analysis device 1. The analyzer 1 executes processing based on the sound data received from the sound data acquisition device 3. In the example of FIG. 5, the sound data acquisition device 3 is attached to the equipment 5, but may be installed around the equipment 5.

FIG. 6 is a functional block diagram of the analyzer 1. The analyzer 1 includes a receiving unit 101, a clustering unit 103, a calculation unit 105, an associating unit 107, an evaluation unit 109, a sound data storage unit 111, a cluster data storage unit 113, and a probability data storage unit 115. And the correspondence data storage unit 117 and the evaluation result storage unit 119.

In the receiving unit 101, the clustering unit 103, the calculating unit 105, the associating unit 107, and the evaluation unit 109, for example, the program loaded in the memory 2501 in FIG. 30 is executed by the CPU (Central Processing Unit) 2503 in FIG. It is realized by. The sound data storage unit 111, the cluster data storage unit 113, the probability data storage unit 115, the correspondence data storage unit 117, and the evaluation result storage unit 119 are provided in, for example, the memory 2501 or the HDD (Hard Disk Drive) 2505 in FIG. ..

The receiving unit 101 receives the sound data from the sound data acquisition device 3, and stores the received sound data in the sound data storage unit 111. The clustering unit 103 executes processing based on the sound data stored in the sound data storage unit 111, and stores the processing result in the cluster data storage unit 113. The calculation unit 105 executes processing based on the data stored in the cluster data storage unit 113, and stores the processing result in the probability data storage unit 115. The association unit 107 executes processing based on the data stored in the sound data storage unit 111, the data stored in the cluster data storage unit 113, and the data stored in the probability data storage unit 115, and the processing result is obtained. Is stored in the correspondence data storage unit 117. The evaluation unit 109 executes processing based on the data stored in the cluster data storage unit 113 and the data stored in the correspondence data storage unit 117, and stores the processing result in the evaluation result storage unit 119.

Next, the details of the process executed by the analyzer 1 of the present embodiment will be described with reference to FIGS. 7 to 26.

FIG. 7 is a diagram showing a processing flow of processing for associating an operation mode with each cluster. This process is executed prior to the process of determining whether or not an abnormal noise is being emitted from the equipment 5.

The clustering unit 103 of the analyzer 1 reads the sound data received from the sound data acquisition device 3 from the sound data storage unit 111 (FIG. 7: step S1). The sound data read in step S1 is, for example, sound wave data as shown in FIG.

The clustering unit 103 extracts the feature amount at each time by spectral decomposition of the sound data read in step S1 (step S3). When the feature amount is represented as a multidimensional vector, time series data of the feature amount is obtained for each dimension in step S3. In the example of FIG. 9, time-series data of the feature amount A, the feature amount B, ... Are obtained. Since the method of spectral decomposition is well known, it will not be described in more detail here.

The clustering unit 103 stores the extracted feature amount in the cluster data storage unit 113. FIG. 10 is a diagram showing an example of data stored in the cluster data storage unit 113. In the example of FIG. 10, the feature amount is stored for each time. Since clustering has not been performed at the time of step S3, the cluster number is not stored.

The clustering unit 103 executes clustering for the feature amount extracted in step S3 to generate three clusters (step S5). The clustering is, for example, k-means clustering or clustering by a mixed normal distribution.

FIG. 11 is a diagram showing an example of the distribution of points corresponding to the feature amount. Each point corresponds to a feature amount at time t (i) (i = 1, 2, ..., N). In such an example, when clustering is executed so as to generate three clusters, clusters 1 to 3 are generated, for example, as shown in FIG.

The clustering unit 103 stores the result of the clustering executed in step S5 in the cluster data storage unit 113. FIG. 13 is a diagram showing an example of data stored in the cluster data storage unit 113. In the example of FIG. 13, the feature amount and the cluster number are stored for each time. The cluster number is the number of the cluster to which the point corresponding to the feature amount at time t (i) belongs.

The calculation unit 105 calculates the number of transitions from each cluster to another cluster using the data stored in the cluster data storage unit 113 (step S9). For example, when clusters transition as shown in FIG. 14, the number of transitions from cluster 1 to cluster 2 is 2, and the number of transitions from cluster 1 to cluster 3 is 4. The number of transitions from cluster 2 to cluster 1 is 3, and the number of transitions from cluster 2 to cluster 3 is 1. The number of transitions from cluster 3 to cluster 1 is 4, and the number of transitions from cluster 3 to cluster 2 is 1.

The calculation unit 105 calculates the probability of transition from each cluster to another cluster using the calculation result of step S9 (step S11). For example, in the case of FIG. 14, the transition probability from cluster 1 to cluster 2 is 2 / (2 + 4) ≈0.33, and the transition probability from cluster 1 to cluster 3 is 4 / (2 + 4) ≈0.67. is there. The transition probability from cluster 2 to cluster 1 is 3 / (3 + 1) = 0.75, and the transition probability from cluster 2 to cluster 3 is 1 / (3 + 1) = 0.25. The transition probability from cluster 3 to cluster 1 is 4 / (4 + 1) = 0.8, and the transition probability from cluster 3 to cluster 2 is 1 / (4 + 1) = 0.2.

The calculation unit 105 stores the transition probability calculated in step S11 in the probability data storage unit 115. FIG. 15 is a diagram showing an example of data stored in the probability data storage unit 115. In the example of FIG. 15, for each cluster, the transition probability to another cluster is stored. At the time of step S11, no value is stored in the transition probability difference column.

The calculation unit 105 calculates the transition probability difference for each cluster using the transition probability stored in the probability data storage unit 115 (step S13), and stores the calculated transition probability difference in the probability data storage unit 115. Then, the process proceeds to step S15 of FIG. 17 via the terminal A.

FIG. 16 is a diagram showing an example of data stored in the probability data storage unit 115. In the example of FIG. 16, for each cluster, the transition probability to another cluster and the transition probability difference are stored.

Moving on to the description of FIG. 17, the association unit 107 identifies the cluster having the smallest transition probability difference among the three transition probability differences (FIG. 17: step S15).

The association unit 107 stores data indicating the non-steady operation mode in the correspondence data storage unit 117 in association with the cluster number specified in step S15 (step S17).

The processing of steps S15 and S17 is based on the characteristics of the cluster transition. FIG. 18 is a diagram showing the transition of clusters. As described above, the transition from the stop mode to the steady operation mode and the transition from the steady operation mode to the stop mode go through the unsteady operation mode, so that the direct transition from the stop mode to the steady operation mode is performed. And in theory, a direct transition from steady-state mode to stop mode cannot occur. Therefore, it is considered that the transition probability from the stop mode to the steady operation mode and the transition probability from the steady operation mode to the stop mode are lower than the other transition probabilities. From this, it is considered that the transition probability difference for the stop mode and the transition probability difference for the steady operation mode are larger than the transition probability difference for the unsteady operation mode. Therefore, the cluster in the unsteady operation mode can be specified by the process of step S15.

FIG. 19 is a diagram showing an example of data stored in the correspondence data storage unit 117. In the example of FIG. 19, the cluster number and the information indicating the operation mode are stored. At the time of step S17, the information indicating the steady operation mode and the information indicating the stop mode are not stored.

The association unit 107 calculates effective values for two clusters other than the cluster specified in step S15 (step S19). The effective value is calculated according to, for example, the following formula.

FIG. 20 is a diagram for explaining an effective value. As described above, since no sound is emitted from the equipment 5 in the stop mode, it is considered that the effective value of the sound wave type in the stop mode is smaller than the effective value of the sound wave type in the steady operation mode. Therefore, for example, in the case of FIG. 20, the cluster corresponding to FIG. 20A is a cluster in stop mode, and the cluster corresponding to FIG. 20B is a cluster in steady operation mode.

The association unit 107 stores data indicating that the steady operation mode is in the correspondence data storage unit 117 in association with the cluster number having the larger effective value calculated in step S19 (step S21). ..

The association unit 107 stores data indicating that the stop mode is in the correspondence data storage unit 117 in association with the cluster number having the smaller effective value calculated in step S19 (step S23). And the process ends. By the processing up to this point, data as shown in FIG. 21, for example, is stored in the correspondence data storage unit 117.

Here, the effect of the above processing will be described.

FIG. 22 is a diagram showing three clusters of features when there is no environmental sound, and FIG. 23 is a diagram showing three clusters of features when there is environmental sound. Then, as shown in FIG. 24, the true operation mode of the cluster corresponding to “◯” is the stop mode, and the true operation mode of the cluster corresponding to “Δ” is the steady operation mode, which is set to “+”. The true mode of operation of the corresponding cluster is the transient mode of operation. The cluster number corresponding to "○" is 1, the cluster number corresponding to "Δ" is 2, and the cluster number corresponding to "+" is 3.

When the association is executed based on the cluster radius, the association can be correctly performed in the example of FIG. 22 where there is no environmental sound. However, in the case of the example of FIG. 23 where there is an environmental sound, since the cluster radius of the cluster corresponding to “Δ” is large, the unsteady operation mode is erroneously associated with the cluster corresponding to “Δ”. There is a possibility that it will end up.

On the other hand, according to the process of the present embodiment, the transition probability difference is calculated as shown in FIG. 25, for example. In the case of the example of FIG. 24, since the transition probability difference of the cluster having the number 3 (that is, the cluster corresponding to “+”) is the smallest, the unsteady operation mode is associated with the cluster corresponding to “+”. be able to. Further, based on the magnitude of the effective value, the steady operation mode is associated with the cluster having the number 2 (that is, the cluster corresponding to "Δ"), and the cluster having the number 1 (that is, "○") is associated with the steady operation mode. The stop mode can be associated with the corresponding cluster). That is, the association is performed as shown in FIG. 26, and this association corresponds to the true operation mode shown in FIG. 24.

Next, a process of determining whether or not an abnormal noise is being emitted from the equipment 5 will be described with reference to FIGS. 27 to 29.

FIG. 27 is a diagram showing a processing flow of a process of determining whether or not an abnormal noise is emitted from the equipment 5. This process is executed after the process described with reference to FIGS. 7 to 26 is executed.

First, the clustering unit 103 reads out the sound data newly stored in the sound data storage unit 111 and newly received from the sound data acquisition device 3 (FIG. 27: step S31). The sound data read in step S31 is sound data different from the sound data in which the processing described with reference to FIGS. 7 to 26 has been executed.

The clustering unit 103 extracts the feature amount at each time by spectral decomposition of the sound data read in step S31 (step S33). The clustering unit 103 passes the extracted feature amount to the evaluation unit 109.

The evaluation unit 109 identifies the cluster number associated with the data indicating the steady operation mode in the correspondence data storage unit 117, and reads out the feature amount of the specified cluster number from the cluster data storage unit 113 (step S35). ..

When there are a plurality of points corresponding to the feature amounts extracted in step S33, the following processes in steps S37 to S43 are executed for each point.

The evaluation unit 109 calculates the degree of abnormality of the feature amount extracted in step S33 by using the feature amount extracted in step S33 and the feature amount read out in step S35 (step S37).

The degree of abnormality will be described with reference to FIG. 28. FIG. 28 is a diagram showing the positional relationship between the three clusters and the points to be evaluated. In FIG. 28, it is assumed that cluster 1 is a cluster in unsteady operation mode, cluster 2 is a cluster in stop mode, and cluster 3 is a cluster in steady operation mode. Point 2801 is located away from the cluster in steady-state mode, and point 2802 is in the cluster in steady-state mode.

In the present embodiment, the points belonging to the clusters in the steady operation mode are present in the space where the points are densely present, and the points belonging to the clusters in the steady operation mode are sparsely present. Is calculated so that the degree of abnormality is high. Therefore, in the example of FIG. 28, the degree of abnormality at point 2801 is higher than the degree of abnormality at point 2802. Since the method of calculating such an abnormality degree is well known, it will not be described in more detail here.

The evaluation unit 109 determines whether the degree of abnormality calculated in step S37 is equal to or greater than the threshold value (step S39). When the degree of abnormality calculated in step S37 is equal to or greater than the threshold value (step S39: Yes route), the evaluation unit 109 executes the following processing. Specifically, the evaluation unit 109 stores the data indicating that the noise is abnormal in the evaluation result storage unit 119 in association with the identification information of the point to be evaluated (step S41).

On the other hand, when the degree of abnormality calculated in step S37 is not equal to or greater than the threshold value (step S39: No route), the evaluation unit 109 executes the following processing. Specifically, the evaluation unit 109 stores data indicating that the sound is normal in the evaluation result storage unit 119 in association with the identification information of the point to be evaluated (step S43). And the process ends.

FIG. 29 is a diagram showing an example of data stored in the evaluation result storage unit 119. In the example of FIG. 29, the identification information of the point to be evaluated, the degree of abnormality, and the data indicating that the sound is abnormal or the data indicating that the sound is normal are stored.

As described above, by using the order characteristic of switching operation modes, it becomes possible to correctly associate the operation modes with each cluster, and as a result, it becomes possible to carry out acoustic diagnosis with high accuracy. .. In particular, when there is an environmental sound, it has not been possible to appropriately associate it with the prior art, but it is possible by the method of the present embodiment.

Although one embodiment of the present invention has been described above, the present invention is not limited thereto. For example, the functional block configuration of the analyzer 1 described above may not match the actual program module configuration.

Further, the configuration of each table described above is an example, and does not have to be the configuration as described above. Further, in the processing flow, the order of processing can be changed as long as the processing result does not change. Further, it may be executed in parallel.

Further, the method of the present embodiment capable of appropriately specifying the operation mode is not a method applicable only to abnormal noise detection, but is applicable to general operation monitoring.

The analyzer 1 described above is a computer device, and as shown in FIG. 30, a drive for a memory 2501, a CPU 2503, an HDD 2505, a display control unit 2507 connected to the display device 2509, and a removable disk 2511. The device 2513, the input device 2515, and the communication control unit 2517 for connecting to the network are connected by a bus 2519. The operating system (OS: Operating System) and the application program for executing the processing in this embodiment are stored in the HDD 2505, and are read from the HDD 2505 to the memory 2501 when executed by the CPU 2503. The CPU 2503 controls the display control unit 2507, the communication control unit 2517, and the drive device 2513 according to the processing contents of the application program to perform a predetermined operation. Further, the data in the process of processing is mainly stored in the memory 2501, but may be stored in the HDD 2505. In an embodiment of the present invention, the application program for performing the above-described processing is stored and distributed on a computer-readable removable disk 2511 and installed from the drive device 2513 to the HDD 2505. It may be installed on the HDD 2505 via a network such as the Internet and a communication control unit 2517. Such a computer device realizes various functions as described above by organically collaborating with the hardware such as the CPU 2503 and the memory 2501 described above and the program such as the OS and the application program. ..

The embodiments of the present invention described above can be summarized as follows.

In the evaluation method according to the first aspect of the present embodiment, (A) the feature amount at each time is extracted from the sound data, and a plurality of feature amount clusters are generated by clustering the extracted feature amount, and (B). Based on the change of the cluster to which the feature belongs at each time, the probability of transition from each of the plurality of clusters to another cluster is calculated, and (C) based on the probability of transition from each of the plurality of clusters to another cluster. For each of the plurality of clusters, the first data indicating that the target is in steady operation, the second data indicating that the target is in non-steady operation, or that the target is stopped. The process of associating the third data to be shown is included.

The characteristic is that the transition from steady operation to stop and the transition from stop to steady operation do not theoretically occur. Therefore, by using the probability of transition between clusters, it becomes possible to correctly associate the operation mode with each cluster of the feature amount extracted from the sound data.

Further, in the process of associating each of the plurality of clusters, (c1) for each of the plurality of clusters, the difference in the probability of transition from the cluster to another cluster is calculated, and the calculated difference is the smallest. The second data may be associated with one cluster.

The cluster for unsteady operation is characterized in that the difference in the probability of transition tends to be smaller than that for the cluster for steady operation and the cluster for stoppage. Therefore, if the above processing is executed, the association for the unsteady operation can be correctly performed.

Further, in the process of associating each of the plurality of clusters, (c2) for each of the clusters other than the first cluster, the effective value of the sound data for the feature amount belonging to the cluster is calculated, and the calculated effective value is calculated. The first data is associated with the second cluster, which is the larger cluster, and the third data is associated with the third cluster, which is the cluster with the smaller effective value. You may.

The effective value for steady operation tends to be larger than the effective value for stoppage. Therefore, if the above processing is executed, the correspondence between the steady operation and the stop can be correctly performed.

Further, in the process of calculating the probability of transition from each of the plurality of clusters to another cluster, (b1) the fourth data including the identification information of the cluster to which the feature amount belongs at each time is generated, and (b2) the second. Transition from each of the plurality of clusters to another cluster by dividing the number of transitions from the cluster to each other cluster by the total number of transitions for the cluster for each of the plurality of clusters based on the data of 4. You may calculate the probability of.

The probability of transition from each cluster to another cluster can be calculated appropriately.

Further, in this evaluation method, when (D) other sound data different from the sound data is acquired, the feature amount is extracted from the other sound data, and (E) the feature amount extracted from the other sound data and the first It may be determined whether or not an abnormal noise has occurred in the target based on the degree of abnormality calculated based on the feature amount belonging to the cluster to which the data of 1 is associated.

It becomes possible to determine with high accuracy whether or not an abnormal noise has occurred in the target.

The evaluation device according to the second aspect of the present embodiment is (E) a generation unit that extracts the feature amount at each time from the sound data and generates a plurality of feature amount clusters by clustering the extracted feature amount (implementation). The clustering unit 103 in the form of is an example of the generation unit) and (F) Calculation to calculate the probability of transition from each of a plurality of clusters to another cluster based on the change of the cluster to which the feature amount belongs at each time. Based on the unit (the calculation unit 105 in the embodiment is an example of the calculation unit) and (G) the probability of transition from each of the plurality of clusters to another cluster, the target is for each of the plurality of clusters. Correspondence unit (implementation) that associates the first data indicating that the target is in steady operation, the second data indicating that the target is in non-steady operation, or the third data indicating that the target is stopped (implementation). The association unit 107 in the form of is an example of the association unit).

A program for causing a computer to execute the processing by the above method can be created, and the program can be a computer-readable storage medium such as a flexible disk, a CD-ROM, a magneto-optical disk, a semiconductor memory, or a hard disk. Stored in storage. The intermediate processing result is temporarily stored in a storage device such as a main memory.

The following additional notes will be further disclosed with respect to the embodiments including the above embodiments.

(Appendix 1)
On the computer
The feature amount at each time is extracted from the sound data, and a plurality of clusters of the feature amount are generated by clustering the extracted feature amount.
Based on the change of the cluster to which the feature quantity belongs at each time, the probability of transition from each of the plurality of clusters to another cluster is calculated.
First data indicating that the target is in steady operation for each of the plurality of clusters based on the probability of transition from each of the plurality of clusters to another cluster, the target is in non-steady operation. The second data indicating that the object is stopped, or the third data indicating that the target is stopped is associated with the data.
An evaluation program that executes processing.

(Appendix 2)
In the process of associating with each of the plurality of clusters
For each of the plurality of clusters, the difference in the probability of transition from the cluster to another cluster is calculated, and the second data is associated with the first cluster having the smallest calculated difference.
The evaluation program described in Appendix 1.

(Appendix 3)
In the process of associating with each of the plurality of clusters
For each of the clusters other than the first cluster, the effective value of the sound data for the feature amount belonging to the cluster is calculated, and the calculated effective value is larger than that of the second cluster. The first data is associated with each other, and the third data is associated with the third cluster, which is the cluster having the smaller effective value calculated.
The evaluation program described in Appendix 2.

(Appendix 4)
In the process of calculating the probability of transition from each of the plurality of clusters to another cluster.
At each time, a fourth data including the identification information of the cluster to which the feature amount belongs is generated.
Based on the fourth data, for each of the plurality of clusters, the number of transitions from the cluster to each of the other clusters is divided by the total number of transitions for the cluster, so that each of the plurality of clusters can be replaced with another. Calculate the probability of transition to a cluster,
The evaluation program according to any one of Appendix 1 to 3.

(Appendix 5)
When other sound data different from the sound data is acquired, the feature amount is extracted from the other sound data, and the feature amount is extracted.
It is determined whether or not an abnormal sound has occurred in the target based on the degree of abnormality calculated based on the feature amount extracted from the other sound data and the feature amount belonging to the cluster to which the first data is associated.
The evaluation program described in Appendix 1.

(Appendix 6)
The computer
The feature amount at each time is extracted from the sound data, and a plurality of clusters of the feature amount are generated by clustering the extracted feature amount.
Based on the change of the cluster to which the feature quantity belongs at each time, the probability of transition from each of the plurality of clusters to another cluster is calculated.
First data indicating that the target is in steady operation for each of the plurality of clusters based on the probability of transition from each of the plurality of clusters to another cluster, the target is in non-steady operation. The second data indicating that the object is stopped, or the third data indicating that the target is stopped is associated with the data.
Evaluation method to execute the process.

(Appendix 7)
A generation unit that extracts features at each time from sound data and generates a plurality of clusters of the features by clustering the extracted features.
A calculation unit that calculates the probability of transition from each of a plurality of clusters to another cluster based on the change in the cluster to which the feature quantity belongs at each time.
First data indicating that the target is in steady operation for each of the plurality of clusters, based on the probability of transition from each of the plurality of clusters to another cluster, the target is in non-steady operation. The associating unit that associates the second data indicating that the target is, or the third data indicating that the target is stopped.
Evaluation device with.

1 Analyzer 101 Reception unit 103 Clustering unit 105 Calculation unit 107 Correspondence unit 109 Evaluation unit 111 Sound data storage unit 113 Cluster data storage unit 115 Probability data storage unit 117 Correspondence data storage unit 119 Evaluation result storage unit

Claims (7)

On the computer
The feature amount at each time is extracted from the sound data, and a plurality of clusters of the feature amount are generated by clustering the extracted feature amount.
Based on the change of the cluster to which the feature quantity belongs at each time, the probability of transition from each of the plurality of clusters to another cluster is calculated.
Based on the probability of transition from each of the plurality of clusters to another cluster, the cluster in which the target is in non-steady operation is identified among the plurality of clusters, and the target is not related to the specified cluster. Correspond with the first data indicating that it is in steady operation,
An evaluation program that executes processing. In the process of associating the first data
For each of the plurality of clusters, the difference in the probability of transition from the cluster to another cluster is calculated.
The first cluster with the smallest calculated difference is identified, and the first data is associated with the identified first cluster.
The evaluation program according to claim 1. On the computer
For each of the clusters other than the first cluster, the effective value of the sound data for the feature amount belonging to the cluster was calculated.
For the second cluster, which is the cluster with the larger calculated effective value ,
The target is stopped with respect to the third cluster, which is the cluster having the smaller effective value calculated by associating the second data indicating that the target is in steady operation. Associate the third data shown,
The evaluation program according to claim 2, wherein the processing is further executed. In the process of calculating the probability of transition from each of the plurality of clusters to another cluster.
At each time, a fourth data including the identification information of the cluster to which the feature amount belongs is generated.
Based on the fourth data, for each of the plurality of clusters, the number of transitions from the cluster to each of the other clusters is divided by the total number of transitions for the cluster, so that each of the plurality of clusters can be replaced with another. Calculate the probability of transition to a cluster,
The evaluation program according to any one of claims 1 to 3. When other sound data different from the sound data is acquired, the feature amount is extracted from the other sound data, and the feature amount is extracted.
It is determined whether or not an abnormal sound has occurred in the target based on the degree of abnormality calculated based on the feature amount extracted from the other sound data and the feature amount belonging to the cluster to which the second data is associated.
The evaluation program according to claim 3, wherein the computer further executes the process. The computer
The feature amount at each time is extracted from the sound data, and a plurality of clusters of the feature amount are generated by clustering the extracted feature amount.
Based on the change of the cluster to which the feature quantity belongs at each time, the probability of transition from each of the plurality of clusters to another cluster is calculated.
Based on the probability of transition from each of the plurality of clusters to another cluster, the cluster in which the target is in non-steady operation is identified among the plurality of clusters, and the target is not related to the specified cluster. Correspond with the first data indicating that it is in steady operation,
Evaluation method to execute the process. A generator that extracts features at each time from sound data and generates a plurality of clusters of the features by clustering the extracted features.
A calculation unit that calculates the probability of transition from each of a plurality of clusters to another cluster based on the change in the cluster to which the feature quantity belongs at each time.
Based on the probability of transition from each of the plurality of clusters to another cluster, the cluster in which the target is in non-steady operation is identified among the plurality of clusters, and the target is not related to the specified cluster. An associating unit that associates the first data indicating that steady operation is in progress,
Evaluation device with.

Images