KR101962558B1

KR101962558B1 - Abnormal sound diagnostic apparatus, abnormal sound diagnostic system, abnormal sound diagnostic method, and abnormal sound diagnostic program

Info

Publication number: KR101962558B1
Application number: KR1020177023765A
Authority: KR
Inventors: 요시하루 아베; 히로시 후쿠나가
Original assignee: 미쓰비시덴키 가부시키가이샤; 미쓰비시 덴키 빌딩 테크노 서비스 가부시키 가이샤
Priority date: 2015-02-03
Filing date: 2015-02-03
Publication date: 2019-03-26
Also published as: DE112015006099B4; JP6250198B2; WO2016125256A1; JPWO2016125256A1; CN107209509A; KR20170108085A; CN107209509B; DE112015006099T5

Abstract

A trajectory feature extraction unit 5 for extracting a trajectory vector 15 by converting a trajectory showing the intensity characteristics in the entire time direction of the strength time series 14 acquired by the strength time series acquisition unit 4 into a vector, And an identification parameter storage section for storing the learned identification parameter (16) by outputting information indicative of a state type of the diagnosis object device, An identification unit 7 for obtaining a K-dimensional score vector 17 for each state type of the diagnostic target apparatus from the locus vector 15 and the identification parameter 16, a K-dimensional score vector 17 And a judging section 8 for judging whether or not the sound generated in the diagnosis target device is normal or abnormal.

Description

Abnormal sound diagnostic apparatus, abnormal sound diagnostic system, abnormal sound diagnostic method, and abnormal sound diagnostic program

The present invention is an abnormality diagnosis apparatus for analyzing sounds generated from a device to be diagnosed and diagnosing the types of abnormal noises and abnormal sounds generated by the device, An abnormal sound diagnostic system, an abnormal sound diagnostic method, and an abnormal sound diagnostic program.

Conventionally, as an abnormal sound diagnostic apparatus, an analysis result of sound data collected in a state in which a device to be diagnosed normally operates is stored as a reference value, and the analysis result of the sound data collected at the time of diagnosis deviates from the stored reference value It is known to diagnose that an abnormality has occurred in the device.

For example, the abnormal sound detection device disclosed in Patent Document 1 detects and stores a negative frequency band that is collected when the elevator is normally operated, and excludes the sound of the stored frequency band from the sound collected during the diagnosis operation Thereby diagnosing the presence of an abnormal sound.

The abnormality diagnosis apparatus disclosed in Patent Document 2 acquires a normal time-frequency distribution as a reference at the time of diagnosis and compares the normal time-frequency distribution with the diagnosis time-frequency portion acquired in the diagnosis mode, And the abnormality is judged by comparing the calculated abnormality with the threshold value.

Japanese Laid-Open Patent Publication No. 1-166935 Japanese Laid-Open Patent Publication No. 2013-200143

However, in the above-described technologies of Patent Documents 1 and 2, in order to diagnose the device, when a sound collector is installed at the same position as that at the time of diagnosis in a device that is normally operating, It has been necessary to analyze the sound generated by the sound collector by analyzing it and to learn the criteria for diagnosing the presence of the abnormal sound in advance.

For this reason, in the case where it is not possible to pick up the sound during normal operation before the diagnosis of the apparatus, for example, in the case of an elevator installed at the time of contract, the criteria for diagnosis can not be prepared, There is a problem that it can not be done.

In addition, as described above, when it is not possible to collect a sound during normal operation and a reference for diagnosis can not be prepared, a sound for normal operation is collected by using another apparatus having the same specification, and a criterion for diagnosis is prepared You can also think of how to do it. However, in the case of a complicated device composed of a large number of parts, the specifications such as the installation position of the sound collector, the size of the components constituting the device, and the arrangement conditions of the device, such as the height of the building, It is not practical to prepare a device that is set to have the same specifications such as size, material of the hoistway, cargo load, speed of operation, etc., and it is difficult to prepare appropriate standards using other devices There was a challenge.

An object of the present invention is to diagnose the operation state of the device without requiring the sound collection during the normal operation in advance for the device to be diagnosed.

The abnormal sound diagnostic apparatus according to the present invention includes a sound collecting section for collecting sounds generated in a diagnosis target device and acquiring sound data, a strength time series acquiring section for acquiring an intensity time series from a time frequency distribution obtained by analyzing waveform data of sound data, A trajectory feature extracting unit for extracting a trajectory vector by converting a trajectory showing the intensity characteristics in the entire time direction of the strength time series into a vector and extracting a trajectory vector from the time frequency distribution obtained by analyzing the waveform data of the sound data generated from the reference device An identification parameter storage section for storing a learned identification parameter as an input of a vector which is a locus indicating a strength characteristic in the entire time direction of the strength time series and outputting information indicating a state type of the diagnosis target device; A score for each state type of the diagnostic target device And a judging section for judging whether or not the sound generated in the diagnostic target apparatus is normal or abnormal with reference to the score.

According to the present invention, it is possible to diagnose the presence or absence of an abnormal sound even for a device which can not collect a sound during normal operation and can not create a criterion for diagnosis.

BRIEF DESCRIPTION OF DRAWINGS FIG. 1 is a block diagram showing the configuration of an abnormal sound diagnostic apparatus according to a first embodiment. FIG.
2 is a diagram showing a configuration of an identification unit of the abnormal sound diagnostic apparatus according to the first embodiment.
3 is a block diagram showing a configuration of an identification parameter learning apparatus according to the first embodiment.
4 is a diagram showing an example of accumulation of a database of an identification parameter learning apparatus according to the first embodiment.
5 is a flowchart showing the operation of the abnormal sound diagnostic apparatus according to the first embodiment.
6 is a flowchart showing the operation of the abnormal sound diagnosis apparatus according to the first embodiment.
7 is a diagram showing an example of an anomaly type and a K-dimension score vector referenced by the determination section of the abnormal-sound diagnostic apparatus according to the first embodiment.
8 is an explanatory diagram showing the effect of the abnormality diagnosis by the abnormality diagnosis apparatus of the first embodiment;
Fig. 9 is an explanatory view showing the result of abnormal sound diagnosis by a conventional abnormal sound diagnosis apparatus; Fig.
10 is an explanatory diagram showing the connection of a plurality of intensity vectors by the trajectory feature extraction unit of the abnormality diagnosis apparatus according to the first embodiment;
11 is an explanatory diagram showing the effect of the abnormality sound diagnosis according to the other constitution of the abnormality diagnosis apparatus of the first embodiment;
12 is a diagram showing a configuration of an identification unit of the abnormal sound diagnosis apparatus according to the second embodiment.
13 is a flowchart showing the operation of the abnormality diagnosis apparatus according to the second embodiment.

BEST MODE FOR CARRYING OUT THE INVENTION Hereinafter, embodiments for carrying out the present invention will be described with reference to the accompanying drawings in order to explain the present invention in more detail.

Embodiment 1

The abnormality diagnosis apparatus according to the first embodiment diagnoses a sound generated from a device (for example, an elevator or the like) to be diagnosed and determines whether or not the generated sound is a normal sound or more, . The device to be diagnosed is a device composed of a plurality of movable parts such as an elevator, for example. The sound collecting means for collecting the generated sound is attached to the inside or outside of the car of the elevator so that the sound generated when the car reciprocates And determines whether the sound collected and collected is normal or abnormal, thereby diagnosing the moving sound of the movable part. Further, the abnormality diagnosis apparatus of the present invention can be applied to other than the elevator.

In the following, a case where the abnormal sound diagnostic apparatus is implemented as software on a personal computer (hereinafter referred to as PC) will be described as an example. The PC has a USB terminal and a LAN terminal, a microphone is connected to the USB terminal through an audio interface circuit, and a device to be diagnosed is connected to the LAN terminal through a LAN cable. The device to be diagnosed is configured to perform a predetermined driving operation by a control instruction output from the PC. Further, the abnormal sound diagnostic apparatus 100 is not limited to the case where it is implemented as software, and can be appropriately changed.

1 is a block diagram showing a configuration of an abnormal-sound diagnostic apparatus 100 according to the first embodiment.

Fig. 1 (a) is a functional block diagram of the abnormal sound diagnosis apparatus 100 according to the first embodiment, which includes a sound collecting section 1, a waveform acquisition section 2, a time frequency analysis section 3, (4), a trajectory feature extraction unit (5), an identification parameter storage unit (6), an identification unit (7), and a determination unit (8).

The sound collector 1 is composed of a sound collector such as a microphone and collects sounds generated from the device to be diagnosed in synchronization with the operation of the device to be diagnosed and outputs sound data 11 . When the device to be diagnosed is an elevator, the sound collecting part 1 is disposed inside the car or outside the car of the elevator. The waveform acquisition section 2 is constituted by, for example, an amplifier and an A / D converter, and samples the waveform of the sound data 11 picked up by the sound pickup section 1 and outputs the waveform data 12, .

The time frequency analysis unit 3 multiplies the waveform data 12 output from the waveform acquisition unit 2 by a time window and performs a fast Fourier transform (hereinafter referred to as FFT) operation while shifting the time window in the time direction, Data 12 is time-frequency analyzed, and time-frequency distribution 13 is obtained. The strength time series acquisition unit 4 obtains an intensity time series 14 indicating the strength with respect to time and frequency from the time frequency distribution 13 output from the time frequency analysis unit 3. [ The trajectory feature extraction unit 5 smoothes the intensity time series 14 output from the strength time series acquisition unit 4 in the time direction and extracts a trajectory vector 15 extending over the entire time axis. The identification parameter storing section 6 is a storage area for storing previously learned identification parameters, and stores an identification parameter for identifying whether the operation state of the device is normal or abnormal, and an identification parameter for identifying the type Remember the parameters. Details of the learning of the identification parameter 16 stored in the identification parameter storage unit 6 will be described later.

The identification unit 7 combines the identification parameter 16 stored in the identification parameter storage unit 6 and the locus vector 15 extracted by the locus feature extraction unit 5 to obtain a score for a plurality of abnormal types do. It is assumed that K abnormal kinds such as a normal operation state and an abnormal operation state at a specific position are set as the abnormal types. The score for the K kinds of abnormal types is hereinafter referred to as a K-dimensional score vector 17. The detailed configuration of the identification unit 7 will be described later. Based on the K-dimensional score vector 17 of the identification unit 7, the determination unit 8 determines whether the operation state of the device is normal or abnormal. If the operation state is abnormal, the determination unit 8 also makes a determination, 18).

Fig. 1B is a block diagram showing a hardware configuration of the abnormal-sound diagnostic apparatus 100 according to the first embodiment, which is composed of a processor 100a and a memory 100b. The waveform shaping unit 2, the time frequency analysis unit 3, the intensity time series acquisition unit 4, the trajectory feature extraction unit 5, the identification unit 7, (100a) executes the program stored in the memory 100b. It is assumed that the identification parameter storage unit 6 is stored in the memory 100b.

Next, the detailed configuration of the identification unit 7 will be described.

2 is an explanatory view showing the configuration of the identification unit 7 of the abnormal-sound diagnostic apparatus 100 according to the first embodiment, and shows a configuration of a neural network in the identification unit 7. Fig.

The neural network shown in the example of Fig. 2 is configured in a hierarchical structure, and is composed of one input layer 71 and two first hidden layers 72 and two hidden layers 73, which are hidden layers. The input layer 71, the first hiding layer 72 and the second hiding layer 73 are provided with a unit for simulating the function of the synapse of the cerebral nerve circuit. There is no coupling between the units in each layer and only the coupling between the units between the layers. For this reason, it is known that the neural network of the first embodiment can stably obtain good performance according to a learning method known as Deep Learning in the field of machine learning.

The last hidden layer also serves as an output layer. In the example of Fig. 2, the second hidden layer 73 also serves as an output layer. In addition, the number M of hidden layers is generally an integer of 1 or more layers (M? 1). Hereinafter, the case where the number of hidden layers M = 2 will be described with reference to FIG. 2 as an example.

The input layer 71 has the same number of units as the number of dimensions (for example, L x B) of the trajectory vector 15 input from the trajectory feature extraction unit 5. [ In addition, the second hidden layer 73, that is, the output layer has K non-linear units equal to the number K of abnormal types. The number of hidden layer units excluding the output layer is set to a predetermined number in consideration of the identification performance of the neural network. (M) (m = 0, 1, 2, ..., M), where 0 is the input layer and m is the number of units in the mth layer, .

In Equation (1), U (m) represents the number of units of the m-th layer.

The load and bias necessary for calculating the response of the hidden layer are supplied from the identification parameter 16 stored in the identification parameter storage unit 6. [ Let w (i, j, m-1) and c (j, m-1) be the load and bias supplied to the mth hidden layer. The range of i and j is i = 0, 1, ..., U (m-1) -1 and j = 0, 1, ..., U (m) -1.

Next, learning of the identification parameter 16 used in the identification unit 7 will be described. The identification parameter 16 stored in the identification parameter storage section 6 is learned by the identification parameter learning apparatus 200 shown in Fig.

3 (a) is a functional block diagram of the identification parameter learning apparatus 200 according to the first embodiment, and shows a functional block diagram of a discrimination parameter learning apparatus 200 according to the first embodiment. The discrimination parameter learning apparatus 200 includes a sound data generation section 21, a sound database 22, a waveform acquisition section 23, A trajectory feature extraction unit 26, a teacher vector creation unit 27,

The sound data generation unit 21 collects sound data using a plurality of devices having different specifications or operations as reference devices, or generates sound data by computer simulation. In the example of the first embodiment, a plurality of elevators having different specifications and operations become reference devices. In the sound database 22, sound data 22a and abnormal type data 22b are stored. The sound data 22a is composed of sound data generated by the sound data generation unit 21 and sound data superimposed on the sound data 22a generated by the sound data generation unit 21. [ The abnormality data 22b includes an abnormality type of the device associated with the sound data 22a, specifically, a label indicating whether the operation state of the device is normal or abnormal, Is stored.

An example of the sound data 22a and the abnormal type data 22b stored in the sound database 22 is shown in Fig.

4, the sound data 22a is composed of "serial number", "object name" and "sound data file name", and the abnormal type data 22b is composed of "abnormal" Type C (v): Yes ".

As an example of the abnormal class C (v), "normal", "abnormal", "abnormal", and "abnormal middle layer" are associated with each other. The above types are set.

The waveform acquisition unit 23 samples the waveform of the sound data 22a stored in the sound database 22 and outputs the waveform data 31 converted into a digital signal. 1, the time frequency analysis unit 24, the intensity time series acquisition unit (parameter intensity time series acquisition unit) 25 and the trajectory feature extraction unit (parameter trajectory feature extraction unit) The time frequency analysis unit 3, the strength time series acquisition unit 4 and the trajectory feature extraction unit 5 of the sound diagnostic apparatus 100 perform the same operations and calculate the time frequency distribution 32, the intensity time series 33, And a trajectory vector 34 are output. The teacher vector creating unit 27 creates the teacher vector 35 by using the abnormal type data 22b stored in the sound database 22.

The identification learning unit 28 creates learning data for learning a neural network. Learning data of a neural network generally consists of input data and output data expected to be output by the neural network when the input data is given. In the example of the block diagram shown in Fig. 3, the input data is the trajectory vector 34 input from the trajectory feature extraction unit 26, and the output data is the teacher vector 35 input from the teacher vector creation unit 27 .

Assuming that the total number of sound data used for learning of the neural network is V, the input data is V trajectory vectors 34 and the output data is V teacher vectors 35. [

The input data x (k, v) in the discrimination learning section 28 is expressed by the following equation (7), where ρ (k, v) is the locus vector 34 extracted from the v- (2).

That is, the input data x (k, v) is the same as the trajectory vector 34.

The V teacher vectors 35 generated by the teacher vector generator 27 are set such that K is the number of kinds of abnormal types, y (k, v) is the kth element of the vth teacher vector, Let y (k, v) be the vector with the C (v) th element as 1 and the other element as 0, where y (k, v) is the ideal kind of sound data.

The identification learning unit 28 performs learning of the neural network by using the trajectory vector 34 which is the input data obtained as described above and the teacher data vector 35 as the output data, And stores the obtained parameter in the identification parameter storage section 6 as the identification parameter 36. [ The load and the bias constituting the identification parameter 36 are the load w (i, j, m-1) used in calculating the response of the first hidden layer 72 and the second hidden layer 73 of the above- And the bias c (j, m-1).

3 (b) is a block diagram showing the hardware configuration of the identification parameter learning apparatus 200 according to the first embodiment, and is composed of a processor 200a and a memory 200b. The waveform data acquisition unit 21, the waveform acquisition unit 23, the time frequency analysis unit 24, the strength time series acquisition unit 25, the trajectory feature extraction unit 26, the teacher vector creation unit 27, (28) is realized by the processor (200a) executing the program stored in the memory (200b). It is assumed that the tone database 22 is stored in the memory 200b.

Next, the operation of the abnormal sound diagnostic apparatus 100 will be described with reference to Figs. 5 and 6. Fig.

5 and 6 are flowcharts showing the operation of the abnormal sound diagnosis apparatus 100 according to the first embodiment. Fig. 5 shows the operation of the sound collecting section 1 and the waveform acquisition section 2, And the operations of the respective components after the frequency analysis section 3 are shown. In the following, a device to be diagnosed by the abnormal-sound diagnostic apparatus 100 will be simply referred to as a device.

When the abnormal sound diagnostic apparatus 100 detects the start of operation of the apparatus (step ST1), the sound collecting section 1 picks up the sound generated from the apparatus (step ST2). The waveform acquisition unit 2 samples and amplifies the sound data 11 picked up in step ST2, amplifies the sound data, performs A / D conversion, and samples a negative waveform (step ST3) Bit linear PCM (pulse code modulation) digital signal waveform data (step ST4).

Next, the abnormal-sound diagnostic apparatus 100 determines whether or not the operation of the apparatus has been terminated (step ST5). If the operation of the device has not been terminated (step ST5; NO), the process returns to step ST2 and the above-described process is repeated. On the other hand, when the operation of the apparatus is ended (step ST5; YES), the waveform acquisition section 2 concatenates the waveform data obtained in step ST4 and outputs it as a series of waveform data 12 (step ST6). Thus, the acquisition processing of the acquisition and waveform data ends. Next, proceeding to the flowchart of Fig. 6, abnormality sound diagnosis processing using the acquired waveform data 12 is performed.

The time frequency analysis unit 3 acquires the waveform data 12 output from the waveform acquisition unit 2 and outputs a time window of, for example, 1024 points to the waveform data 12 at intervals of 16 ms in the time direction And a time frequency distribution g (t, f), which is a series of frequency spectrums, is obtained by an FFT calculation for each frame to obtain a time frequency distribution 13 (step ST11).

Here, t is the time index corresponding to the shift interval for shifting the time window, and f is an index indicating the frequency of the result of the FFT operation. The time t and the frequency f are integers satisfying 0? T? T and 0? F? F, respectively. T is the number of frames in the time direction of the time frequency distribution 13 and F is an index corresponding to a nyquist frequency that is 1/2 of the sampling frequency fs of the waveform data 12 (F = fs / 2).

Next, the intensity time series obtaining unit 4 obtains the time-frequency distribution 13 obtained by the step ST11 using five frequency bands of, for example, 0.5 kHz, 1 kHz, 2 kHz, 4 kHz, and 8 kHz as center frequencies, The sum of the frequency components included in the five frequency bands is obtained, and the intensity time series 14 of each band is obtained (step ST12). Assuming that the intensity time series 14 of each band is G (t, b), G (t, b) is given by the following equation (4).

In Expression (4), b is an index of a band, and is an integer satisfying 0? B <B (where B is the number of bands, B = 5 in this example). Further, Ω (b) represents a set of frequencies f to be searched for the band b in the time-frequency distribution g (t, f).

The trajectory feature extraction unit 5 smoothes the intensity time series 14 in the time direction for each band (step ST13), obtains the smoothing intensity at the point where the entire time axis is divided into L parts, and creates an L intensity vector (Step ST14). In this example, the intensity time series 14 is smoothed in the time direction in five bands. The intensity is further normalized with respect to the L-dimensional intensity vector created in step ST15, and the L-dimensional intensity vector of each band after the normalization is concatenated to generate an LXB-dimensional locus vector 15 (step ST16).

The smoothed time series G ~ (t, b) (t = 0, 1, ..., T-1, b = 0, 1, ..., B- .

In equation (5), smooth_t (x (t)) is a function that outputs a new time series obtained by smoothing series x (t) related to t in the subscript t direction.

The smoothing intensity H (l, b) (l = 0, 1, ..., L-1, b = 0, 1, ..., B-1) of the L- Is calculated based on equation (6).

In Equation (6), τ (l) is a real number function indicating an interpolation position related to the subscript t in G ~ (t, b), and w (l) is a function for assigning a weighting coefficient in interpolation. (7) and (8).

Int (x) in the expression (8) is a function for obtaining the integer part of the argument x.

In generating the locus vector 15 in step ST16, the smoothing intensity H (l, b) of the L equally divided points of each band is concatenated as an L-dimensional vector to obtain a locus vector 15, (k) is given by the following equation (9), assuming that the kth element is p (k) (k = 0, 1, ..., L x B-1).

Next, the identification unit 7 inputs the trajectory vector 15 input from the trajectory feature extraction unit 5 to the input layer 71 of the neural network, and outputs the identification parameter stored in the identification parameter storage unit 6 as To calculate the activity of the output unit, and generates the K-dimensional score vector 17 (step ST17).

The processing in step ST17 will be described with reference to a specific configuration example of the identification unit 7 in Fig. First, the i-th element in the locus vector 15 is copied to the i-th unit of the input layer. Assuming that the value of the i-th unit of the input layer is x (i, 0), x (i, 0) is given by the following equation (10).

In Expression (10), p (i) represents the value of the i-th element of the trajectory vector 15.

Next, the outputs of the respective units are sequentially calculated from the first hiding layer 72 to the second hiding layer 73. The output of each unit is obtained by multiplying the output from all units in the preceding layer by the load, summing the sum, subtracting the bias, and performing the nonlinear transformation with the sigmoid function. Assuming that the output of the jth unit of the mth hidden layer is x (j, m), x (j, m) is calculated by the following equation (11).

In Equation (11),? (X) is a sigmoid function having a nonlinear input / output characteristic representing the threshold characteristic of soft, and is given by the following Expression (12).

In the above-described equation (11), when m = 1, x (i, 0) is required. This is because the i-th element ρ i.

By performing the calculation based on the equation (11) up to m = 1, ..., M, the output x (k, M) of the last hidden layer is obtained. In the example of FIG. 2, the output x (k, 2) of the second hidden layer 73 is obtained. The corresponding output is regarded as the output of the output layer. Assuming that the kth output of the output layer is o (k), o (k) is given by the following equation (13).

Finally, the K outputs of the output layer are normalized. By normalization, the total sum of the K outputs becomes 1. Assuming that the result of the normalization is the value s (k) of the score vector, the value s (k) of the score vector is given by the following equation (14) known as the softmax operation.

The K-dimensional score vector 17 obtained by the above-described processing is output to the determination section 8. [

Returning to the flowchart of Fig. 6, description will be continued.

The determining unit 8 compares the elements of the K-dimensional score vector 17 generated in step ST17 and determines the possible types of abnormalities based on the index of the largest element (step ST18), outputs the determination result ( Step ST19), and the process is terminated. If there is a possibility that the type is k *, then k * is given by the following equation (15).

In addition, although a configuration is shown in which the score of the K-dimensional score vector 17 outputs one maximum element, a plurality of elements may be output together with their scores.

7 is a diagram showing an example of an abnormal type and a K-dimensional score vector referenced by the determination section 8 of the abnormal-sound diagnostic apparatus 100 according to the first embodiment.

As shown in Fig. 7, K " abnormal types " are associated with " K dimension score vectors ", respectively. The K-dimensional score vector becomes " 1 " by adding all the values of the K score vectors constituting the K-dimensional score vector. In the example of Fig. 7, since the score vector of the abnormal type "abnormal part" takes a maximum value of "0.64", the judging section 8 judges that the possible abnormal type is "abnormal part".

Next, the effect when the abnormality diagnosis apparatus 100 configured as described above is applied to an elevator will be described with reference to Figs. 8 and 9. Fig.

8 is an explanatory diagram showing the effect of the abnormal sound diagnosis by the abnormal sound diagnosis apparatus 100 according to the first embodiment. In addition, as a comparison, FIG. 9 shows the result of the abnormal sound diagnosis by the conventional abnormal sound diagnosis apparatus.

First, the method of the abnormal sound diagnosis by the conventional abnormal sound diagnostic apparatus and the obtained result will be described with reference to Fig. In the conventional abnormality diagnosis apparatus, the running section 301 of the car 300 is divided and the negative signal strength occurring at the normal time is stored as a reference value for each divided section. In the example of Fig. 9 (a), the running section is divided into six parts, and the first reference value, the second reference value, ..., the sixth reference value are acquired and stored.

By comparing the stored reference value with the intensity time series of the sound data obtained at the time of diagnosis, the above detection was performed in each section. Since the negative signal strength at the normal time in each section differs depending on the use or the operating environment of each elevator, the reference value obtained for an elevator can not be applied to the diagnosis of an abnormal sound of an elevator, There is a problem that the accuracy of the abnormality diagnosis is lowered. Therefore, in the conventional abnormality diagnosis apparatus, it is necessary to perform the learning operation in advance for each elevator, and to store the reference value.

Fig. 9 (b) is a diagram prepared based on the negative signal strength at the time of diagnosis when the reference value created in different elevators is applied to another elevator and the result of the comparison is compared with the negative signal strength at the time of diagnosis. There is an entity 304 of the operation unless there is a normal operation entity 303 whose signal strength exceeds the reference value 305 or the signal strength does not exceed the reference value 305. [ Thus, even when the reference value 305 is set, there is a problem that the normal operation state of the device can not be clearly separated from the abnormal operation state based on the negative signal strength at the time of diagnosis.

Next, the effect of the abnormality sound diagnosis by the abnormality sound diagnosis apparatus 100 according to the first embodiment will be described with reference to FIG.

In the abnormal sound diagnosis apparatus 100 according to the first embodiment, as shown in Fig. 8 (a), the sound generated while the car 300 travels reciprocally between the lowermost layer and the uppermost layer is picked up, Analysis is performed, a strength time series is obtained, and a locus vector is extracted by vector-converting the locus over the entire length in the time direction of the strength time series as a whole. In the example of Fig. 8 (a), in order to simplify the explanation, it is assumed that the abnormal type is two kinds of "normal" and "abnormal" (K = 0 to 1) And the L x 1-dimensional trajectory vectors 306 and 307 are extracted. The trajectory vector 306 indicates a vector when the abnormal type is "1: abnormal", and the locus vector 307 indicates a vector when the abnormal type is "0: normal". When the locus vector 306 and the locus vector 307 are input to the identification unit 7, the identification unit 7 obtains the result of plotting the position in space of the locus vector 306 and the locus vector 307 8 (b).

Fig. 8 (b) is a graph showing the relationship between the first characteristic axis (main axis) and the second characteristic axis (axis perpendicular to the main axis) And the arrangement of each vector is shown on an L x 1 -dimensional space formed by the characteristic axes thereof.

Principal component analysis is a process for displaying the positional relationship of mutualities in a multidimensional space of a vector, and is not a process constituting the present invention. The first characteristic axis and the second characteristic axis are not calculated by the configuration of the present invention, but are described to indicate that the locus vector is classified in space.

In the case where the group 308 representing the apparatus's normal state and the group 309 representing the apparatus malfunction are arranged based on the abnormal type of the trajectory vector and the spatial position of the trajectory vector as shown in the plot result of Figure 8 (b) (Straight line) orthogonal to the straight line connecting the center of the circle 308 and the center of the circle 309 is obtained as the boundary 310. [ With this arrangement, we can understand the general characteristics of the strength time series.

8 (b) shows an example in which a straight line is obtained as the boundary 310, but in actual diagnosis processing, it is assumed that a super-curved line (curve) having a complicated shape is obtained.

As described above, it is possible to grasp the general characteristics represented by the strength time series, which do not depend on the elevator specifications and the operating environment, so that it is not necessary to learn the reference values for each individual in advance and the robustness of the elevator specifications and operating environment ) Diagnosis can be performed.

As described above, according to the first embodiment, the sound collecting unit 1 for picking up the sound generated from the device, the waveform acquiring unit 2 for acquiring the waveform data obtained by sampling the waveform of the collected sound data, A time-frequency analysis unit (3) for performing time-frequency analysis of the acquired waveform data, an intensity time series acquisition unit (4) for obtaining an intensity time series representing the intensity with respect to time and frequency from the time-frequency distribution, A trajectory feature extraction unit 5 for smoothing in the time direction and extracting a trajectory vector over the entire time axis, and a trajectory feature extraction unit 5 that uses the extracted trajectory vector as input data, An identification unit (7) for acquiring a score vector corresponding to the abnormal type based on the locus vector and the stored identification parameter; It is possible to grasp the generalized characteristics indicated by the intensity time series, which do not depend on the specifications of the device and the operating environment, so as to determine whether the device is normal or abnormal, It is possible to determine an abnormal type. Thereby, it is not necessary to learn the reference at the time of diagnosis for each apparatus having different specification or operation in advance, and it is possible to perform robust diagnosis even for the difference of the specifications of the apparatus and the operating environment. It is also possible to provide an abnormal-sound diagnostic apparatus that suppresses deterioration of diagnostic accuracy due to differences in specifications of devices and operating environments.

In the above description of the first embodiment, the sound pickup unit 1 is constituted by a single sound pickup unit and arranged in a device to be diagnosed. However, the sound pickup unit 1 may be constituted by a plurality of sound pickup units, It may be arranged at a plurality of places of the device. In this case, the multi-channel sound data 11 is obtained in synchronization with the operation of the device to be diagnosed, and simultaneously, the multi-channel sound data 11 is obtained. The waveform acquisition section 2, the time frequency analysis section 3 and the strength time series acquisition section 4 acquire the waveform data 12, the time frequency distribution 13 and the intensity time series 14 Conduct. The trajectory feature extraction unit 5 acquires the intensity vectors of the multiple channels from the multichannel strength time series 14 input from the strength time series acquisition unit 4. [ Also, the intensity vectors of the respective channels are connected in the time axis direction.

10 is an explanatory diagram showing the connection of intensity vectors of multiple channels in the trajectory feature extraction unit 5 of the abnormal-sound diagnostic apparatus 100 according to the first embodiment.

10 shows a case of connecting the intensity vectors of three channels and connects the vector 15a of the first channel, the vector 15b of the second channel and the vector 15c of the third channel in the time axis direction of the vector Thereby generating a trajectory vector 15 of L x B x three dimensions (" x 3 " depends on the concatenation of the intensity vectors of the three channels). Since the connections between the channels are in the middle layer of the neural network, it is possible to learn the synchronicity between the channels. In the description up to the preceding paragraph, the number of dimensions of the trajectory vector is L x B, but here, the number of dimensions of the trajectory vector is replaced by L x B x 3.

As described above, by using sound data collected by a plurality of sound collectors, the degree of separation in the identification space between vectors having different types of abnormalities is improved, and the diagnostic accuracy can be improved.

Fig. 11 is an explanatory diagram showing the effect of the abnormality diagnosis based on a locus vector obtained by connecting intensity vectors of multiple channels. Fig.

In FIG. 11A, the intensity time series 311, 312, and 313 represent intensity time series obtained in the first frequency band, the second frequency band, and the third frequency band, respectively, and the intensity time series 311, 312, Dimensional trajectory vector 314 and trajectory vector 315, which are obtained by connecting the vectors obtained from the trajectory vectors in the direction of the time axis. The locus vector 314 indicates a vector when the abnormal type is "1: abnormal", and the locus vector 315 indicates a vector when the abnormal type is "0: normal". When the trajectory vector 314 and the trajectory vector 315 are input to the identification unit 7, the result obtained by plotting the positions of the trajectory vector 314 and the trajectory vector 315 in space in the identification unit 7 is Is shown in Fig. 11 (b). In Fig. 11 (b), the same results as those shown in Fig. 8 (b) can be obtained.

Embodiment 2 Fig.

In the first embodiment described above, the description has been given of the case where the identification unit 7 is a neural network configuration. In the second embodiment, a support vector machine (hereinafter referred to as SVM) is applied as the identification unit .

Since the overall configuration of the abnormal-sound diagnostic apparatus 100 according to the second embodiment is the same as that of the first embodiment, description of the block diagram will be omitted and the identification unit having a different configuration will be described in detail below.

12 is a diagram showing a configuration of an identification section 7a of the abnormal-sound diagnosis apparatus 100 according to the second embodiment.

The identification unit 7a has (K-1) K / 2 SVMs as a whole when K is the number of abnormal types. Here, each SVM is constructed so as to classify and identify any two or more kinds of vectors among K abnormal classes including normal. Each SVM has a number n of support vectors, n support vectors x _i (i = 0, 1, 2, ..., n-1), n coefficients α _i (i = 0, 1, 2,. ..., n-1), a bias b, and a definition k (x ₁ , x ₂ ) of a kernel function to be described later. SVM [i, j] (0? I <j <K) describes the SVM that identifies the normal or abnormal type i and the abnormal type j (where i <j).

Next, the operation of the identification unit 7a will be described.

13 is a flowchart showing the operation of the abnormality diagnosis apparatus according to the second embodiment. Hereinafter, the same steps as those of the abnormality diagnosis apparatus according to the first embodiment will be assigned the same reference numerals as those used in Fig. 6, and the description thereof will be omitted or simplified. The operations of the collector section 1 and the waveform acquisition section 2 are the same as those of the flowchart shown in Fig. 5 of the first embodiment, and a description thereof will be omitted.

When the locus vector 15 generated by the locus feature extraction unit 5 is inputted in step ST16, the identification unit 7a inputs the locus vector 15 to each SVM and stores it in the identification parameter storage unit 6 The output value y (?) Of the identification function of each SVM is calculated based on the following equation (16) (step ST21).

Here, the inner product <φ (x1), φ ( x2)> between k (x _1, x ₂₎ are mapped to a multi-dimensional space of vectors x1 φ (x ₁₎ and the vector map of x2 in the multidimensional space of φ (x2) (Where φ (x) is a nonlinear function of vector x that can not be expressed by an explicit expression). As the kernel function, for example, a Gauss kernel expressed by the following equation (17) can be used. Also,? Is a parameter of the Gaussian kernel.

Next, the identification unit 7a calculates the classification output of each class from the output values of the identification functions of the respective SVMs calculated in step ST21, and calculates the value s of the score vector indicating the scores of 1 to K corresponding to the abnormal types (k), and the calculated value s (k) of the score vector is output to the determination section 8 as the K-dimension score vector 17 (step ST22). The judging unit 8 compares the elements of the K-dimension score vector 17 generated in step ST17 and judges a possible abnormal type based on the index of the largest element (step ST18), outputs the judgment result ( Step ST19), and the process is terminated.

As described above, even when the support vector machine is applied to the identification unit 7a as in the second embodiment, the trajectory feature extraction unit 5 can obtain the trajectory vector 15 (FIG. 15) over the entire length in the time direction of the intensity time series 14 ), And it is possible to grasp a generalized characteristic represented by the intensity time series, which does not depend on the specifications of the apparatus or the operating environment. As a result, it is not necessary to learn the criterion at the time of diagnosis for each individual in advance, and a robust diagnosis can be made for the difference of the specifications of the apparatus and the operating environment. In addition, it is possible to provide an abnormal-sound diagnostic apparatus in which deterioration of the diagnostic accuracy due to the difference of the apparatus is suppressed.

In the first and second embodiments described above, the trajectory vector 15 output by the trajectory feature extraction unit 5 is a feature of the trajectory over the entire length in the time direction of the intensity time series 14 by the linear interpolation An L-dimensional vector may be obtained by using another transformation that preserves the characteristics of the trajectory over the entire length in the time direction of the intensity time series 14 regardless of the loss of the information . As other transforms, for example, the locus over the entire length in the time direction of the intensity time series 14 may be Fourier transformed, and an L-dimensional vector may be formed from the Fourier coefficients of a lower order. Alternatively, the compressed feature may be output as an L-dimensional vector by principal component analysis as other transforms.

In addition, the above-described lossless transformation shows that the vector representing the characteristic over the entire length in the time direction of the intensity time series 14 is not subjected to any processing, and the characteristic as it is is used as a vector. On the other hand, the conversion allowing the loss is a process for reducing the number of dimensions by multiplying, for example, a matrix obtained by principal component analysis, on a vector representing a characteristic over the entire length in the time direction of the intensity time series 14 And then the feature is used as a vector. It is considered that a part of the information included in the original feature vector is lost due to the reduction processing of the above-mentioned dimension number.

In the above-described first and second embodiments, when the device to be diagnosed is an elevator, the trajectory over the entire length in the time direction of the intensity time series is set as a vector However, it is also possible to divide the trajectory vector into a section for each one way such as a rising section and a falling section during the reciprocating operation of the moving section, and for each divided section, a trajectory over the entire length in the time direction of the section as a vector And the locus vector is extracted, and the identifying section 7 may be provided for each divided section to carry out the identification processing.

As a result, in the case of an elevator, there is no abnormality at the time of rise, and even if there is an abnormality at the time of descent, diagnosis can be performed for each section.

Further, the segment to be divided may be divided not only into the rising section and the falling section but also the rising section, for example, into a more detailed section such as a lower section, a middle section, and a higher section.

(Industrial availability)

The abnormal sound diagnostic apparatus according to the present invention can be applied to an apparatus which can not generate a reference value for determining an abnormal sound for each individual because the abnormal sound diagnostic apparatus can perform the abnormal sound diagnosis with high accuracy with respect to the specification of the apparatus or the difference in operation, It is suitable for diagnosing the abnormal sound of the device.

1: collector part 2: waveform acquisition part
3: time frequency analysis unit 4: intensity time series acquisition unit
5: locus feature extraction unit 6: identification parameter storage unit
7, 7a: Identification part 8:
21: sound data generating unit 22: sound database
22a: sound data 22b: abnormal type data
23: waveform acquisition unit 24: time frequency analysis unit
25: Strength Time Series Acquisition Unit 26: Trajectory Feature Extraction Unit
27: teacher vector creating unit 28: identification learning unit
71: input layer 72: first hiding layer
73: second hidden layer 100: abnormal sound diagnostic apparatus
100a, 200a: Processor 100b, 200b: Memory
200: identification parameter learning device

Claims

An abnormality sound diagnosing apparatus for diagnosing whether a sound generated in a diagnosis target apparatus is abnormal,
A sound collecting part for collecting sounds generated in the diagnosis target device and acquiring sound data;
A strength time series acquisition unit for acquiring an intensity time series from a time frequency distribution obtained by analyzing waveform data of sound data acquired by the sound collector,
A trajectory feature extracting unit for extracting a trajectory vector by converting a trajectory indicative of an intensity characteristic in the entire time direction of the strength time series acquired by the strength time series acquisition unit into a vector,
A vector which is a trajectory showing a strength characteristic in the whole time direction of the intensity time series acquired from the time frequency distribution obtained by analyzing the waveform data of the sound data generated from the reference device is inputted and information indicating the state classification of the diagnosis target device An identification parameter storing section for storing the learned identification parameter as an output;
An identification unit that obtains a score for each state type of the diagnosis object device from the locus vector extracted by the locus feature extraction unit and the identification parameter stored in the identification parameter storage unit;
And a judging section for judging whether or not the sound generated in the diagnosis target device is normal or abnormal,
The abnormal sound diagnosis apparatus comprising:

The method according to claim 1,
Wherein the strength time series acquisition unit acquires the strength with respect to time and frequency from the time frequency distribution as the strength time series,
The trajectory feature extraction unit converts the locus indicated by the strength time series acquired by the strength time series acquisition unit into a vector in a two-dimensional space with respect to time and frequency, extracts the trajectory vector by connecting the converted vectors
Wherein the abnormal sound diagnostic apparatus comprises:

The method according to claim 1,
Wherein the trajectory characteristic extraction unit performs a lossless vector conversion on the intensity time series acquired by the strength time series acquisition unit or performs vector conversion of loss loss.

The method according to claim 1,
Wherein the identification unit acquires the score using a method of a neural network.

The method according to claim 1,
Wherein the identification unit acquires the score using a method of a support vector machine.

The method according to claim 1,
A plurality of sound collecting units for collecting sound data of a plurality of channels,
Wherein the strength time series acquisition section acquires the intensity time series of the plurality of channels from a time frequency distribution obtained by analyzing waveform data of each sound data of a plurality of channels collected by the collector section,
Wherein the trajectory feature extraction unit is configured to convert the locus indicating the intensity characteristic of the entire time direction of the intensity time series of the plurality of channels acquired by the strength time series acquisition unit into a vector, to connect the converted vectors of the plurality of channels in the time direction, To extract
Wherein the abnormal sound diagnostic apparatus comprises:

The method according to claim 1,
Wherein the diagnosis subject device travels in a movable section composed of a plurality of divided sections,
Wherein the trajectory feature extraction unit extracts the trajectory vector by converting a trajectory indicative of an intensity characteristic of the entire time direction of the strength time series to a vector for each divided section,
The identification unit performs identification processing for each division section
Wherein the abnormal sound diagnostic apparatus comprises:

The abnormality diagnosis apparatus according to claim 1,
Identification parameter learning device
And,
Wherein said identification parameter learning device comprises:
A sound database that stores sound data generated from the reference device, joint noise superimposition data in which sounds are superimposed on the sound data, and abnormality type information of a device associated with the sound data and the joint noise data,
A parameter strength time series acquisition unit for acquiring a strength time series from the time frequency distribution obtained by analyzing the waveform data of the sound data and the joint overlap data accumulated in the sound database,
A parameter trajectory feature extraction unit for converting, from a strength time series acquired by the parameter strength time series acquisition unit, a trajectory indicative of an intensity characteristic in the entire time direction of the strength time series to a vector,
A teacher vector creating unit that creates a teacher vector from the abnormal type information stored in the sound database;
Wherein the learning unit performs learning such that the trajectory vector converted by the parameter trajectory feature extraction unit is input so that the teacher vector generated by the teacher vector creation unit becomes an output and stores the result of the learning as an identification parameter of the abnormality diagnosis apparatus The identification learning unit
Wherein the abnormal sound diagnosis system comprises:

An abnormality sound diagnosis method for diagnosing whether or not a sound generated for a device to be diagnosed is abnormal,
A step of collecting sound generated in the diagnostic target device by the collector part to acquire sound data,
A step of acquiring an intensity time series from a time frequency distribution obtained by analyzing waveform data of the sound data,
A step of extracting a trajectory vector by converting a trajectory characteristic section extracting section, which represents a characteristic feature of intensity in the entire time direction of the intensity time series, into a vector,
The identification unit is configured to receive, as an input, a vector which is a locus indicating a strength characteristic in the entire time direction of the intensity time series acquired from the time frequency distribution obtained by analyzing the waveform data of sound data generated from the reference device, A step of obtaining a score for each state type of the diagnostic target device from the trained vector,
The judgment section refers to the score, and judges whether or not the sound generated in the diagnosis target device is normal or abnormal,
Wherein the abnormal tone diagnosis method comprises the steps of:

A sound collecting processing procedure for collecting sounds generated in the diagnosis target device and acquiring sound data,
A strength time series acquisition processing procedure for acquiring an intensity time series from a time frequency distribution obtained by analyzing the waveform data of the sound data,
A trajectory feature extraction processing procedure for extracting a trajectory vector by converting a trajectory showing the intensity characteristic of the entire time direction of the intensity time series to a vector,
A vector which is a trajectory showing a strength characteristic in the whole time direction of the intensity time series acquired from the time frequency distribution obtained by analyzing the waveform data of the sound data generated from the reference device is inputted and information indicating the state classification of the diagnosis target device An identification processing procedure for obtaining a score for each state type of the diagnostic target device from the learned identification parameter and the locus vector as an output,
Refers to the score, and judges whether or not the sound generated in the diagnostic target device is normal or abnormal,
Readable recording medium storing an abnormal sound diagnostic program for causing a computer to execute an abnormal sound diagnostic program.