Abnormal diagnosis device, abnormal diagnosis method, and manufacturing method of electrical equipment

Abstract

An abnormality diagnosing device (100) comprises a sound collector (2) that converts a sound generated by a determination target object (1) into an analog signal, a signal converter (3) that converts the analog signal into a digital signal, and a signal processing device (10), wherein: the signal processing device (10) includes a noise determination unit (4) that calculates a feature quantity for each time interval of the digital signal and uses the feature quantity to determine the presence or absence of noise contamination for each time interval, and an abnormality determining unit (5) that determines the presence or absence of an abnormality in the sound of the determination target object (1) using the digital signal for a time interval not contaminated with noise, as output from the noise determination unit (4); and the feature quantity used for the noise determination is a value indicating a spectral intensity for each time interval in waveform data of the digital signal.

Description

Abnormality diagnosis device and abnormality diagnosis method

[0001] The present disclosure relates to an abnormality diagnosis device and an abnormality diagnosis method for diagnosing abnormalities in products such as electrical equipment.

[0002] In a typical electrical equipment assembly process, an operational test of the electrical equipment is performed during or after assembly is completed. Furthermore, during the operational test, a sensory test is also conducted to check whether the vibrations or operational sounds generated by the electrical equipment are normal or normal. Since such sensory tests are performed using the worker's hearing or tactile sense, they are characterized by being dependent on the worker's senses. Therefore, in order to quantify the vibration or operational sound information used to determine whether the electrical equipment is normal or normal, diagnostic devices that convert information about vibrations or operational sounds acquired by microphones, vibration sensors, etc. into waveform data and process the signals are sometimes used. However, since the data acquired using these methods contains vibrations or sounds other than those of the electrical equipment, it is necessary to remove these vibrations and sounds from the acquired data before performing an abnormality diagnosis.

[0003] Patent Document 1 listed below discloses a technology for determining whether or not noise is mixed in by using a drive sound collector installed near a product to collect the drive sound of the product and a noise collector installed at a position away from the product to collect ambient noise. In Patent Document 1 listed below, it is determined that noise is mixed in when the difference between the sound data of the drive sound collector and the sound data of the noise collector or the sound of the noise collector exceeds a preset threshold.

[0004] Japanese Patent Application Laid-Open No. 2006-126141

[0005] As described above, the technology of Patent Document 1 determines whether noise is present by calculating the difference between sound data from two sound collectors located at different distances from the product. Therefore, if the source of the ambient sound is closer to the drive sound collector than to the noise collector, the ambient sound may be mistakenly determined to be the drive sound. This poses a risk of overlooking or falsely detecting an abnormality in the product. Furthermore, the need to prepare multiple sound collectors poses a problem of many constraints, such as the selection of installation locations and equipment.

[0006] The present disclosure has been made to solve the above-mentioned problems, and aims to provide an abnormality diagnosis device that can accurately determine abnormalities in a product using a single sound pickup device.

[0007] In order to solve the above-mentioned problems and achieve the object, the abnormality diagnosis device according to the present disclosure includes a sound collector that collects sound generated by an object to be evaluated and converts it into an analog electrical signal, a signal converter that converts the analog electrical signal into a digital signal, and a signal processing device that performs signal processing on the digital signal, wherein the signal processing device has: a noise determination unit that calculates a feature amount for each time interval of the digital signal based on the digital signal output from the signal converter and performs noise determination using the feature amount to determine whether or not noise is mixed in for each time interval; and an abnormality determination unit that performs abnormality determination using the digital signal output from the noise determination unit for a time interval in which no noise is mixed, to determine whether or not an abnormality exists in the sound of the object to be evaluated, wherein the feature amount used for noise determination is a value that indicates the spectral intensity for each time interval in the waveform data of the digital signal.

[0008] The abnormality diagnosis device according to the present disclosure has the effect of enabling abnormality diagnosis that accurately determines abnormalities in a product using a single sound pickup device.

[0009] FIG. 1 is a block diagram showing a configuration example of an abnormality diagnosis device according to the first embodiment; FIG. 2 is a diagram for explaining a feature vector calculated by a noise determination unit in the abnormality diagnosis device according to the first embodiment; FIG. 3 is a diagram for explaining a feature vector when the values ​​of the shift amount and the time width are equal in the abnormality diagnosis device according to the first embodiment; FIG. 1 is a diagram showing a method for determining a time interval where noise is mixed in the abnormality diagnosis device according to embodiment 1. FIG. 2 is an explanatory diagram for supplementary explanation of a method for determining a time interval where noise is mixed in the abnormality diagnosis device according to embodiment 1. FIG. 3 is a flowchart showing the processing flow of a noise determination unit in the abnormality diagnosis device according to embodiment 1. FIG. 4 is an explanatory diagram showing a method for determining a time interval where noise is mixed in the abnormality diagnosis device according to embodiment 2. FIG. 5 is a diagram for explaining a non-determination target interval in the abnormality diagnosis device according to embodiment 2. FIG. 6 is an explanatory diagram showing a method for determining a time interval where noise is mixed in the abnormality diagnosis device according to embodiment 3.

[0010] An abnormality diagnosis device and an abnormality diagnosis method according to an embodiment of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0011] Embodiment 1. FIG. 1 is a block diagram showing an example of the configuration of an abnormality diagnosis device according to embodiment 1. The abnormality diagnosis device 100 according to embodiment 1 determines whether there is an abnormality in the vibration or operating sound generated from a determination object 1. The determination object 1 is sometimes called a product. The determination object 1 according to embodiment 1 can be a device that has the characteristic of generating periodic and continuous sound or vibration. In this specification, a motor may be used as an example. However, a motor is one example of the determination object 1 and is not limited to this.

[0012] 1 , the abnormality diagnosis device 100 includes a microphone 2, a signal converter 3, and a signal processing device 10. The microphone 2 is sometimes called a sound collector. The signal processing device 10 includes a noise determination unit 4, an abnormality determination unit 5, and a storage unit 6.

[0013] The microphone 2 is placed at a distance from the object to be determined 1. In other words, the microphone 2 and the object to be determined 1 are placed apart from each other and are not in contact with each other. The microphone 2 picks up sounds including the operating sounds of the object to be determined 1 and converts the picked up sounds into analog electrical signals.

[0014] The signal converter 3 is an analog to digital (AD) converter, and converts the analog electrical signal output from the microphone 2 into a digital signal E.

[0015] The signal processing device 10 receives the digital signal E output from the signal converter 3 and performs signal processing, which will be described later. As described above, the signal processing device 10 includes the noise determination unit 4, the abnormality determination unit 5, and the storage unit 6. Note that the abnormality diagnosis device 100 may include a plurality of signal processing devices 10. The signal processing device 10 may further include a display device (not shown) such as a display on which the noise determination results and the abnormality determination results are displayed. The signal processing device 10 may also include a user interface (not shown) that allows the user to perform operations such as input operations. The user interface may include a keyboard, a mouse, a touch panel, etc.

[0016] In the signal processing device 10, the noise determination unit 4 receives the digital signal E output from the signal converter 3 as an input signal. The digital signal E is continuous time-series data. Here, the case where the digital signal E is waveform data is taken as an example. The noise determination unit 4 calculates a feature value of the digital signal E for each time interval based on the digital signal E, and determines whether or not noise is present for each time interval using the feature value. Specifically, the noise determination unit 4 performs necessary arithmetic processing on the waveform data of the input digital signal E to calculate a feature value for each time interval used for noise determination, and determines the time interval in which noise is present based on the feature value. The feature value used for noise determination is, for example, a value indicating the spectral intensity or the amount of variation in the spectral intensity for each time interval in the waveform data of the digital signal E. Based on the result of the noise determination, the noise determination unit 4 transmits the digital signal E for the time interval, excluding the time interval determined to be contaminated with noise, from the waveform data of the input signal, to the abnormality determination unit 5.

[0017] The abnormality determination unit 5 determines whether there is an abnormality in the operating sound of the object 1 by using the digital signal for the time interval received from the noise determination unit 4. That is, the abnormality determination unit 5 performs abnormality determination to determine whether there is an abnormality in the operating sound of the object 1 by using the digital signal for the time interval excluding the time interval for which the noise determination unit 4 has determined that noise is mixed in, i.e., the digital signal for the time interval for which no noise is mixed in.

[0018] The storage unit 6 stores the results of the noise determination by the noise determination unit 4 and the results of the abnormality determination by the abnormality determination unit 5 .

[0019] While a general-purpose microphone can be used as the microphone 2, it is desirable to minimize the inclusion of ambient noise and the like in the operating sounds generated by the object to be determined 1. This can be achieved by using a sound-collecting microphone with sound collection directionality, a parabolic sound-collecting microphone that combines a general microphone with a parabolic reflector, or an appropriate combination of these sound-collecting microphones. Note that the microphone 2 may be a device other than a microphone as long as it is a sound collector that can collect sounds including the operating sounds of the object to be determined 1.

[0020] A general audio interface can be used for the signal converter 3. Here, using an audio interface with a high sampling frequency is desirable because it allows a wide frequency range to be set when performing Short Time Fast Fourier Transform (STFT) processing in the signal processing of the noise determination unit 4 described below. Note that if the signal processing device 10 has a function equivalent to the signal converter 3, that function may be used to perform analog-to-digital conversion.

[0021] The signal processing device 10 may be configured, for example, by a general-purpose computer or a programmable logic controller (PLC), and the processing of the signal processing device 10 may be performed using such devices. PLCs are also called sequencers. The signal processing device 10 may also be configured using an electronic board configured with a signal processing device and a storage medium. Examples of the signal processing device include a microcomputer and an FPGA (Field Programmable Gate Array). Examples of the storage medium include non-volatile or volatile semiconductor memory such as RAM (Random Access Memory), flash memory, EPROM (Erasable Programmable ROM), and EEPROM (Electrically EPROM). The signal processing device 10 may perform a series of operations, from capturing waveform data of the digital signal E to noise and anomaly determination, according to procedural processing. The signal processing device 10 may also have a screen capable of displaying and instructing the status and conditions of noise and anomaly determination. In this case, the screen of the signal processing device 10 may be configured, for example, by a touch panel. In this way, the noise determination unit 4 and the abnormality determination unit 5 of the signal processing device 10 are realized by a processing circuit. The processing circuit may be a processor that executes a program and a memory that stores the program, or it may be dedicated hardware. The processing circuit is also called a control circuit. When the processing circuit is composed of a processor and a memory, each function of the processing circuit is realized by software, firmware, or a combination of software and firmware. The software or firmware is written as a program and stored in the memory. In the processing circuit, each function is realized by the processor reading and executing the program stored in the memory.

[0022] Next, the algorithm of the noise determination process in the signal processing device 10 will be described. The noise determination unit 4 performs STFT processing on the waveform data of the digital signal E converted by the signal converter 3, dividing the waveform data into an arbitrary number of parts in both the frequency and time directions. STFT is a time-frequency analysis technique and is generally known for its ability to perform frequency analysis for each local time. In other words, STFT is capable of performing frequency analysis for each arbitrary time frame. Note that the processing here is not limited to STFT processing, and other time-frequency analysis techniques such as continuous wavelet transform (CWT) may also be used. In the first embodiment, the case where STFT processing is used will be described as an example, and the processing result of STFT is expressed as a matrix F having two dimensions, i frequencies and j times, as shown in the following equation (1).

[0023]

[0024] In the above formula (1), i is the number of divisions of the waveform data in the frequency direction, and j is the number of divisions of the waveform data in the time direction. In this way, when STFT processing is performed, the waveform data of the digital signal output from the signal converter 3 is divided into i frequency frames and j time frames, and the spectral intensity for each can be calculated. Each element of the matrix F represents the corresponding spectral intensity.

[0025] Furthermore, based on matrix F, which is the processing result of the STFT, the noise determination unit 4 calculates the sum of the spectral intensities of one or more frequency frames included in matrix F for each time interval divided by a time width including at least one time frame included in matrix F. Note that here, the sum of the spectral intensities of all frequency frames included in matrix F is calculated. This sum is called a feature. The noise determination unit 4 performs noise determination for each time interval using the feature, determining whether or not noise is present. Specifically, the noise determination unit 4 calculates a feature vector Sn, with the feature for each time interval thus obtained as each element of the vector. The feature vector Sn is given by the following formula (2). In the following formula (2), the "n" in "Sn" is the number n in the time direction of matrix F (hereinafter referred to as time width n) for calculating the sum of the spectral intensities, and t is the number of elements of the feature vector Sn. Note that the time width n may be the same as or different from the time width of the j time frames in the STFT. The time width n can be set to any value as long as it is a natural number satisfying 1≦n. The time span n includes one or more of the j time frames. In addition, although an example in which the sum of the spectral intensities of all frequency frames of the matrix F is calculated as a feature is given here, the sum may be the sum of the spectral intensities of one or more frequency frames included in the matrix F.

[0026]

[0027] FIG. 2 is a diagram illustrating a feature vector calculated by the noise determination unit in the abnormality diagnosis device according to the first embodiment. FIG. 2 shows the relationship between the time width n and the feature vector Sn. In FIG. 2, FIG. 2(a) shows the matrix F of the above equation (1) calculated by the noise determination unit 4 through STFT processing. FIG. 2(b) shows a state in which the j time frames of the matrix F are divided by the time width n, and the time intervals of the time width n are sequentially shifted by a shift amount a. FIG. 2(c) shows the feature vector Sn of the above equation (2) calculated by the noise determination unit 4.

[0028] In the first embodiment, as shown in Fig. 2, the sum of the spectral intensities for each time interval of time width n is calculated by shifting each time interval in the time direction by a shift amount a. In this case, the number of elements t of a feature vector Sn, which has the sum of the spectral intensities for each time interval in matrix F as its feature, can be expressed by the following equation (3).

[0029]

[0030] In the above equation (3), mod(A, B) represents the remainder of A modulo B. For example, the value of mod(7, 2) is 1. Here, the shift amount a and the time width n are natural numbers, and the relationship between them is 1≦a≦n. A time interval r (hereinafter referred to as a non-detection interval r) that is not used to calculate the feature vector Sn, whose feature is the sum of the spectral intensities for each time interval in the matrix F, cannot be subjected to the noise detection process described below. Therefore, the non-detection interval r is not subjected to noise detection, and is treated as a "time interval containing noise" in the subsequent anomaly detection process. Note that r can be calculated by r = mod(j-(n-a), a).

[0031] In this way, each element Sn of the feature vector Sn １ , Sn ２ , ..., Sn ｔ is the sum of the spectral intensities of the i frequencies in the matrix F in a state where time intervals of time width n partially overlap each other.

[0032] In the first embodiment, for ease of understanding, the noise determination process will be described below using an example in which the shift amount a and the time width n are equal. Fig. 3 is a diagram for explaining the feature vector when the shift amount and the time width are equal in the abnormality diagnosis device according to the first embodiment. Fig. 3 shows the relationship between the time width n and the feature vector Sn when the shift amount a and the time width n are equal. In Fig. 3, because the shift amount a and the time width n are equal in value, there is no overlap between the time intervals of the time width n.

[0033] 3, Fig. 3(a) shows the matrix F of the above equation (1) calculated by the noise determination unit 4 through STFT processing. Fig. 3(b) shows a state in which j time frames of the matrix F are divided by a time width n, and the time sections of the time width n are sequentially shifted by a shift amount a (=n). Fig. 3(c) shows the feature vector Sn of the above equation (2) calculated by the noise determination unit 4.

[0034] In the case of Fig. 3, the number of elements t of a feature vector Sn, which has the sum of spectral intensities for each time interval in the matrix F as a feature, can be expressed by the following formula (4). The following formula (4) can be obtained by substituting a = n into the above formula (3).

[0035]

[0036] As described above, in the first embodiment, the noise determination unit 4 performs STFT processing on the sound captured by the microphone 2, and then calculates the sum of the spectral intensities for each fixed time width n. Then, as shown in Fig. 6 (described later), threshold determination using the first threshold value Zn is performed on the sum to identify time periods in which noise is present. The threshold determination using the first threshold value Zn will be described below with reference to Figs. 4 to 6.

[0037] Fig. 4 is an example of an STFT contour diagram showing the spectrum of an operating sound generated by an object to be determined in the abnormality diagnosis device according to embodiment 1. Fig. 4 shows the STFT spectrum intensity of the motor operating sound when the object to be determined 1 is a motor and the motor is being driven. In Fig. 4, the spectrum of the motor operating sound is shown using a figure filled with a pattern. As shown in Fig. 4, the motor operating sound is characterized by a periodic and continuous spectrum.

[0038] On the other hand, Fig. 5 is an example of an STFT contour diagram showing the spectrum of noise in the abnormality diagnosis device according to embodiment 1. In Fig. 5, the spectrum of noise is shown using a figure filled with a pattern. As shown in Fig. 5, the noise is characterized by being aperiodic or sudden.

[0039] In this way, the operating sound of a motor is a periodic and continuous sound, whereas noise is a non-periodic and sudden sound, and this is the point that is focused on in embodiment 1. As described above, when the sum of the spectral intensities is calculated after performing frequency analysis such as STFT, the sum of the spectral intensities in the time intervals where noise is mixed is greater than the sum of the spectral intensities in the time intervals where noise is not mixed, as shown in Figure 6.

[0040] 6A and 6B are examples of STFT contour diagrams showing both the spectrum of the operating sound and the spectrum of noise of the object to be determined in the abnormality diagnosis device according to embodiment 1. Fig. 6A is an image diagram showing the calculation of the sum of the spectral intensities using STFT processing when noise is mixed into the operating sound of the motor. As shown in Fig. 6A, the sound picked up by microphone 2 may contain noise mixed into the operating sound of the motor.

[0041] When the sum of the spectral intensities is calculated over a certain time interval n, the spectral intensities related to the motor operation noise are added to the sum for all time intervals. In Figure 6(b), the sum of the spectral intensities related to the motor operation noise is shown by the white bars. The white bars are not all the same height, but they do not suddenly increase in size. In this way, the spectral intensities related to the motor operation noise are not always constant and there is some variation, but they are generally stable.

[0042] On the other hand, the spectral intensity of noise sounds is added to the sum only in time intervals where noise is present. Therefore, the sum of spectral intensities in time intervals where noise is present is larger than that in time intervals where noise is not present. In Figure 6(b), the sum of spectral intensities in time intervals where noise is present is shown by hatched bars. Here, by setting a first threshold value Zn (described later), it is possible to determine that a time interval in which the sum of spectral intensities is equal to or greater than the first threshold value Zn is a time interval in which noise is present.

[0043] Note that the sum of the spectral intensities may be calculated for all i frequency frames of the STFT, or some frequency frames may be excluded from the calculation of the sum of the spectral intensities among all i frequency frames. Examples of frequency frames that are excluded include a frequency band that is not subject to anomaly detection or the frequency band of a worker's voice.

[0044] 7 is an explanatory diagram showing a method for determining a time interval containing noise in the abnormality diagnosis device according to embodiment 1. Fig. 7 is an image diagram of a method for determining a noise interval using a feature vector Sn with a time width n and a shift amount a, in which the sum of the spectral intensities for each time interval in matrix F is used as a feature. Here, t is the number of elements in the feature vector Sn.

[0045] 7, Fig. 7(a) shows the matrix F of the above equation (1) calculated by the noise determination unit 4 through STFT processing. Fig. 7(b) shows a state in which the j time frames of the matrix F are divided by a time width n, and the time sections of the time width n are sequentially shifted by a shift amount a. Fig. 7(c) shows the feature vector Sn of the above equation (2) calculated by the noise determination unit 4.

[0046] 7D, Hn denotes a noise determination vector. The noise determination vector Hn is a vector obtained by dividing each element Sn １ ~Sn ｔ It has elements corresponding to Sn ｔ The element of Hn corresponding to ａ×（ｔ－１）＋１ ~hn ａ×ｔ Therefore, Sn １ The element of Hn corresponding to １ ~hn ａ are a elements of Sn ２ The element of Hn corresponding to ａ＋１ ~hn ２ａ are a elements of

[0047] Element Sn of feature vector Sn ｔ If the value of is equal to or greater than the first threshold value Zn, the element hn of the noise determination vector Hn ａ×（ｔ－１）＋１ From hn ａ×ｔIn other words, among the elements of the noise determination vector Hn, an element having a value of 1 indicates a time interval in which noise is mixed. On the other hand, the element Sn of the feature vector Sn ｔ If the value of is less than the first threshold value Zn, the element hn of the noise determination vector Hn ａ×（ｔ－１）＋１ From hn ａ×ｔ 0 is input to a elements up to t. That is, elements of the noise determination vector Hn to which 0 is input indicate time intervals in which noise is not mixed. Furthermore, 1 is input to elements of the noise determination vector Hn corresponding to the non-determination interval r'. In the case of FIG. 7, the non-determination interval r' can be calculated by r'=j-a×t.

[0048] FIG. 7 will be described using FIG. 8 as a specific example. FIG. 8 is a diagram for supplementary explanation of the method for determining a time interval containing noise in the abnormality diagnosis device according to embodiment 1. In FIG. 8, the number of elements in the time direction of matrix F is j=21. Furthermore, the time width n of feature quantity vector Sn, which uses the sum of spectral intensities for each time interval in matrix F as a feature quantity, is set to 4, and the shift amount a is set to 4. In this case, the number of elements t of feature quantity vector S4 is 5 according to the above formula (4), and the non-determination interval r'' is 1.

[0049] 8, Fig. 8(a) shows the matrix F of the above equation (1) calculated by the noise determination unit 4 through STFT processing. Fig. 8(b) shows the feature vector Sn (where n = 4) of the above equation (2) calculated by the noise determination unit 4. Fig. 8(c) shows the noise determination vector Hn (where n = 4).

[0050] As shown in FIG. 8, element S4 of feature vector S4 １ is equal to or greater than the first threshold Z4, element S4 １ Four elements h4 of the noise determination vector Hn corresponding to １ , h4 ２ , h4 ３ , h4 ４ On the other hand, element S4 is set to 1. ２ is less than the first threshold Z4, element h4 of the noise determination vector Hn ５ , h4６ , h4 ７ , h4 ８ 0 is input to

[0051] Similarly, each element S4 of the feature vector S4 ｔ For (t=1, ..., 5), a threshold determination of the time interval is performed using the first threshold Z4, and elements S4 that are equal to or greater than the first threshold Z4 are ｔ 1 is input to the element of H4 corresponding to ｔ 0 is input to the element of H4 corresponding to the non-determination section r''. ２１ Enter 1 in

[0052] The first threshold value Zn is a threshold value used to determine a time period in which noise is present from the feature vector Sn calculated in the above process. ｔ On the other hand, for Sn which is less than the first threshold Zn, the corresponding time interval is determined to be a time interval containing noise. ｔ For the feature vector Sn, the corresponding time interval is determined to be a time interval free of noise. The first threshold value Zn is calculated by statistical processing such as the mean value, standard deviation, median, or quartile deviation obtained from the calculated feature vector Sn, or by an appropriate combination of these statistical processing methods.

[0053] Electrical equipment, including motors, varies from product to product. Therefore, even if the noise is the same, a loud operating unit will not affect abnormality diagnosis, but a quiet operating unit may affect abnormality diagnosis. Surrounding environmental sounds also vary in loudness and type. Therefore, the appropriate value for the first threshold value Zn varies depending on the combination of the product sound and the surrounding environmental sound contained in the acquired sound. Therefore, by performing statistical processing on each piece of data and appropriately setting the first threshold value Zn to an appropriate value, appropriate noise removal tailored to the individual differences between products and the surrounding environmental sound is possible.

[0054] In the first embodiment, the first threshold value Zn expressed by the following formula (5) is used. In formula (5), med is the median of the feature vector Sn, iql is the quartile deviation of the feature vector Sn, and coef is a coefficient. Here, the coefficient coef can be set to any value according to the volume level of the noise to be removed. However, the first threshold value Zn is not limited to the example of formula (5).

[0055]

[0056] As described above, in the abnormality diagnosis device 100 according to the first embodiment, after performing STFT processing on the sound acquired by the microphone 2, the sum of the spectral intensities is calculated for each time interval of a certain time width n. This sum is called a feature. The feature vector Sn thus obtained is subjected to threshold determination using a first threshold Zn to determine which time intervals contain noise. In the first embodiment, attention is focused on the fact that motor sound is characterized by periodic and continuous sound, as described in FIG. 4 , whereas noise is characterized by non-periodic and sudden sound, as described in FIG. 5 . When the sum of the spectral intensities is calculated after frequency analysis such as STFT as shown in the first embodiment, the sum of the spectral intensities is large in time intervals containing noise, as described in FIG. 6 . In this manner, in the first embodiment, it is possible to determine and remove time intervals containing a large amount of noise from a single sound data acquired by the microphone 2. As a result, it is possible to determine whether or not there is an abnormality in the sound of the object to be evaluated 1 using only the spectral intensity data for the time period in which no noise is mixed, making it possible to accurately determine whether or not there is an abnormality in the object to be evaluated 1.

[0057] 9 is a flowchart showing the flow of processing by the noise determination unit in the abnormality diagnosis device according to embodiment 1. As shown in Fig. 9, in step S1, the noise determination unit 4 performs STFT processing on the digital signal output from the signal converter 3 to calculate a matrix F having two dimensions, frequency and time.

[0058] Next, in step S2, the noise determination unit 4 calculates, as a feature, the sum of the spectral intensities for each time interval in the matrix F. The feature is expressed as a feature vector Sn.

[0059] Next, in step S3, the noise determination unit 4 determines, based on the feature obtained in step S2, time intervals in the waveform data of the digital signal output from the signal converter 3 that contain noise and time intervals that do not contain noise.

[0060] Next, in step S4, the noise determination unit 4 outputs the time period in which no noise is mixed to the abnormality determination unit 5, and the process ends.

[0061] Next, the abnormality determination unit 5 in the signal processing device 10 will be described. The abnormality determination unit 5 uses data from a time period in which the noise determination unit 4 has determined that no noise is present to perform an abnormality determination for the operating sound of the motor, which is the object of determination 1. This makes it possible to reduce erroneous abnormality determinations due to the influence of noise. Here, a known abnormality determination algorithm can be used for the abnormality determination.

[0062] As described above, the conventional technology described in Patent Document 1 determines whether noise is present by calculating the difference in sound data from two sound collectors located at different distances from the object to be determined. Therefore, the conventional technology described in Patent Document 1 has a problem in that if the noise source is closer to the sound collector that measures the operation sound of the object to be determined than the sound collector that measures the noise, the noise may be erroneously determined to be the operation sound of the object to be determined. Furthermore, the conventional technology described in Patent Document 1 requires the preparation of multiple sound collectors, which results in many constraints, such as the selection of installation locations and equipment.

[0063] In contrast, the abnormality diagnosis device 100 according to the first embodiment uses the above-described algorithm to determine, from sound data collected using a single microphone 2, time periods in which the spectral intensity is equal to or greater than a threshold as time periods containing noise. Therefore, the abnormality diagnosis device 100 according to the first embodiment can identify time periods containing noise regardless of the location of the noise source, and can determine abnormalities in operation sounds using time periods in which noise is not present. This reduces erroneous determinations of abnormalities in operation sounds of the object to be determined 1 due to the influence of noise. As a result, it is also possible to reduce overlooking abnormalities in operation sounds of the object to be determined 1. Furthermore, because the abnormality diagnosis device 100 according to the first embodiment uses only a single microphone 2, it is possible to reduce constraints on the installation location, selection of equipment, and the like.

[0064] Embodiment 2. In the algorithm configuration of the noise determination unit 4 according to the first embodiment, a time interval containing noise is determined using a single noise determination within a certain time interval n, which may result in incomplete noise removal. For example, when a short time interval n is used, it is easy to determine that noise is present in a noise containing locally loud sound. However, when a long time interval n is used, the spectral intensity due to the noise is averaged, which may result in an erroneous determination that noise is not present. On the other hand, when a short time interval n is used, the difference in the sum of the spectral intensities between the short time interval n and other time intervals is small, which may result in an erroneous determination that noise is not present. On the other hand, when a long time interval n is used, the difference in the sum of the spectral intensities between the short time interval n and other time intervals is large, even for quiet noise, compared to a time interval without noise, making it possible to determine that noise is present. Therefore, in the second embodiment, a method that improves on these issues is proposed. The method according to the second embodiment can be implemented using the configuration of the abnormality diagnosis device 100 according to the first embodiment shown in FIG. 1. Therefore, a description of the configuration of the abnormality diagnosis device 100 is omitted here.

[0065] The algorithm of the noise determination unit 4 used in the second embodiment will be described below.

[0066] Based on the matrix F, which is the result of the STFT processing, the noise determination unit 4 calculates a feature vector Sn, the feature of which is the sum of the spectral intensities for each time interval in the matrix F. ｔ For the time intervals, a time interval equal to or greater than a first threshold value Zn is determined to be a time interval containing noise. The processing up to this point is the same as that in the first embodiment. In the second embodiment, the noise determination unit 4 repeats the determination of whether or not a time interval contains noise a number of times by changing the time width n, and determines the number of times that it is determined that noise is contained in the same time frame (i.e., the same time frame out of j time frames). Then, it determines whether or not the determined number of times is equal to or greater than a second threshold value K that is set in advance. Note that the above-described repetitive processing may be performed by changing the shift amount a instead of the time width n.

[0067] In the second embodiment, an example will be described in which the time width n is changed in sequence from n=1, n=2, n=3, ..., n=10, and noise determination is repeated a total of 10 times. Also, in the second embodiment, an example will be described in which the shift amount a and the time width n are equal in value. Below, the algorithm of the noise determination unit 4 used in the second embodiment will be described for this setting. Note that, although the second embodiment uses n=1, ..., n=10 as the time width n, this is not limiting. The time width n may be a combination of any natural numbers. The shift amount a and the time width n may also be different values.

[0068] 10 is an explanatory diagram showing a method for determining a time interval in which noise is present in the abnormality diagnosis device according to embodiment 2. Fig. 10 shows a method for determining a time interval in which noise is present, using noise determination vectors H1 to H10 when the time width n is changed from 1 to 10.

[0069] The elements of the noise determination vector H1 are h1 １ , h1 ２ , ..., h1 ｊThe elements of the noise determination vector H2 are h2 １ , h2 ２ , ..., h2 ｊ Similarly, the elements of the noise determination vector H10 are h10 １ , h10 ２ , ..., h10 ｊ If noise is present, 1 is input to each element of the noise determination vector Hn, and if noise is not present, 0 is input.

[0070] Here, an example will be described in which the number of elements in the time direction of matrix F is j=21. ｊ From h10 ｊ For each element j, the sum of the values ​​of the elements of the noise determination vector Hn is calculated. That is, first, for j=1, the sum of the values ​​of the elements h1 in H1 to H10 is calculated. １ From h10 １ Next, for j=2, calculate the sum of the element values ​​for each element h1 in H1 to H10. ２ From h10 ２ Similarly, for j=3, j=4, ..., j=21, calculate the sum of the values ​​of the elements h1 to H10. ｊ From h10 ｊ The sum of the element values ​​for H1 to H10 is calculated in this way. The result of calculating the sum of each of the 21 elements in H1 to H10 is stored in a total decision value vector G, which has the same number of elements as the number of elements in the time direction (j = 21) in matrix F. In this way, a value between 0 and 10 is input to each element of the total decision value vector G. That is, even if the time width n is changed from 1 to 10 and repeated 10 times, if there is a time frame in which noise has never been mixed in, the value of the element of the total decision value vector G will be 0. On the other hand, if the time width n is changed from 1 to 10 and repeated 10 times, and as a result of the time width n being changed from 1 to 10, and there is a time frame in which it is determined that noise has been mixed in every time, the value of the element of the total decision value vector G will be 10.

[0071] 10, (a) of FIG. 10 shows noise judgment vectors H1 to H10, and (b) of FIG. 10 shows the total judgment value vector G.

[0072] The noise determination unit 4 determines whether each element of the total judgment value vector G is equal to or greater than a second threshold value K. The noise determination unit 4 then determines, among the elements of the total judgment value vector G, time intervals in which the element value is equal to or greater than a predetermined second threshold value K as time intervals in which noise is present. Furthermore, the noise determination unit 4 also sets, among the elements of the total judgment value vector G, time intervals that are not subject to determination as time intervals in which noise is present. The second threshold value K can be a natural number equal to or less than the number of iterations. Therefore, in the second embodiment, since the number of iterations is 10, the second threshold value K can be set to any value between 1 and 10.

[0073] In Figure 10 (b), for example, if the second threshold K is set to K = 7, among the elements of the total judgment value vector G, the time intervals that are greater than or equal to the second threshold K are three: j = 4, where the element value is "8", j = 9, where the element value is "7", and j = 11, where the element value is "8".

[0074] FIG. 11 is a diagram illustrating a non-determination section in the abnormality diagnosis device according to the second embodiment. FIG. 11 shows the relationship between the time width n and the non-determination section r'. FIG. 11 illustrates the non-determination section r' for each of the time widths n = 1, n = 2, ..., n = 10. Here, n = a and j = 21, so similar to the non-determination section r shown in FIG. 3, the non-determination section r' can be calculated as r' = mod (21, n). Therefore, when the time width n = 1, the non-determination section r' is r' = 0; when the time width n = 2, the non-determination section r' is r' = 1; and when the time width n = 10, the non-determination section r' is r' = 1.

[0075] In the second embodiment, the sum of the 21 elements H1 to H10 of the noise judgment vector Hn is calculated to calculate the total judgment value vector G, so that the time intervals in which all 21 elements H1 to H10 are not present must be determined as non-judgment intervals. Therefore, in the second embodiment, the non-judgment intervals are defined as the maximum value of the non-judgment intervals r' in H1 to H10. In the example of FIG. 11, when the time width n is n=8, the non-judgment interval r' has a maximum value of 5. Therefore, the elements g corresponding to j=17 to j=21 of the total judgment value vector G １７ From g ２１ The five time intervals up to are the intervals not subject to judgment.

[0076] The processing flow of the noise determination unit 4 in the second embodiment is basically the same as the flow shown in FIG. 9 in the first embodiment. The difference is that in the second embodiment, the time width n is changed and the noise determination is repeatedly performed to calculate the total determination value vector G. That is, in the second embodiment, in step S3 of FIG. 9 , the noise determination unit 4 distinguishes, based on the feature vector Sn, between time intervals containing noise and time intervals containing no noise in the waveform data of the digital signal output from the signal converter 3, and repeats this distinction multiple times by changing the time width n of the time interval. Then, based on the results of the repeated distinctions, the noise determination unit 4 calculates the total determination value vector G and determines the time interval including the time at which the number of times that noise has been determined to be present, stored in each element of the total determination value vector G, is equal to or greater than the second threshold K. Since the other steps are the same, their description will be omitted here.

[0077] Next, the abnormality determination unit 5 in the signal processing device 10 according to the second embodiment will be described. In the first embodiment described above, abnormality determination of the operation sound is performed using data for a time period in which it is determined by the noise determination vector Hn that no noise is present. In the second embodiment, abnormality determination of the operation sound is performed using data for a time period in which it is determined that no noise is present by threshold determination using the second threshold K in the total determination value vector G. Here, each element of the total determination value vector G contains the number of times it is determined that noise is present in the same time frame in multiple noise determination vectors Hn. This makes it possible to reduce erroneous determination of abnormality due to the influence of noise compared to the first embodiment. Here, a known abnormality determination algorithm can be used for the abnormality determination.

[0078] As described above, the algorithm of the noise determination unit 4 according to the first embodiment determines a time period in which noise is present using a single result for a certain time period n. However, since some noise cannot be detected depending on the set time period, the abnormality diagnosis device 100 according to the first embodiment may erroneously determine an abnormality due to the influence of noise.

[0079] In contrast, the fault diagnosis device 100 according to the second embodiment uses a plurality of time widths to determine the time interval in which noise is present, thereby enabling the time interval in which noise is present to be identified more accurately than the algorithm according to the first embodiment. As a result, the fault diagnosis device 100 according to the second embodiment can reduce erroneous determination of an abnormality due to the influence of noise.

[0080] Embodiment 3. In the algorithm configuration of the noise determination unit according to Embodiments 1 and 2, whether noise is present is determined using the difference between the sum of spectral intensities between a time interval where noise is present and a time interval where noise is not present. Therefore, if noise is present continuously throughout the entire time interval of the acquired sound data, it may be erroneously determined that no noise is present. Therefore, in Embodiment 3, a method that improves on this is proposed. Note that the method according to Embodiment 3 can be implemented using the configuration of the abnormality diagnosis device 100 according to Embodiment 1 shown in FIG. 1.

[0081] The noise determination algorithm used in the third embodiment will be described below.

[0082] In the third embodiment, before carrying out the first or second embodiment, the digital signal E of the waveform data converted by the signal converter 3 is subjected to the processing described below, thereby making it possible to determine the time period in which noise is present even when noise is continuously present throughout the entire time period of the acquired sound data.

[0083] 12 is an explanatory diagram showing a method for determining a time interval in which noise is present in the abnormality diagnosis device according to the third embodiment. FIG. 12 shows a digital signal E of waveform data converted by the signal converter 3, and a signal E' obtained by taking the absolute values ​​of each element of the digital signal E. Here, the number of elements in the digital signal E and the signal E' is y, and elements in the signal E' that are shaded with a pattern indicate time intervals in which the element value is equal to or greater than a preset third threshold value L1. The third threshold value L1 can be set to any value according to the volume level of the noise to be removed.

[0084] Here, a continuous interval is defined as an interval in signal E' in which elements (i.e., time intervals) determined not to contain noise are consecutive. In the example of FIG. 12 , there are four continuous intervals, from continuous interval C1 to continuous interval C4. In embodiment 3, embodiment 1 or embodiment 2 is performed for continuous intervals in which the number of elements included in the continuous interval is equal to or greater than a predetermined fourth threshold value L2. For example, in the example of FIG. 12 , when the value of the fourth threshold value L2 is L2=3, noise determination according to embodiment 1 and embodiment 2 is performed using continuous intervals C3 and C4 from continuous intervals C1 to C4.

[0085] Fig. 13 is a flowchart showing the flow of processing by the noise determination unit in the abnormality diagnosis device according to embodiment 3. As shown in Fig. 13, in embodiment 3, step S0 is performed as a pre-processing step before performing the processing of steps S1 to S4 shown in embodiments 1 and 2. The processing of step S0 will be described below.

[0086] 13, step S0 includes three steps: step S0-1, step S0-2, and step S0-3. However, step S0-3 does not necessarily have to be included in step S0, and may be included in step S0 as necessary.

[0087] In step S0-1, first, the noise determination unit 4 calculates the absolute value of the digital signal E output from the signal converter 3. Hereinafter, the signal obtained by taking the absolute value of each element (i.e., time interval) of the digital signal E will be referred to as signal E'. Specifically, in step S0-1, the noise determination unit 4 calculates the absolute value of the spectral intensity for each time interval included in the digital signal E to generate signal E'.

[0088] Next, in step S0-2, the noise determination unit 4 determines time intervals in which noise is not present, based on a signal E', which is the absolute value of the digital signal E. Specifically, the noise determination unit 4 determines, for each time interval included in the signal E', whether the value of the signal E' is equal to or greater than a third threshold L1. The noise determination unit 4 then determines, as a time interval in which the value of the signal E' is equal to or greater than the third threshold L1, the time interval in which noise is present.

[0089] Next, in step S0-3, the noise determination unit 4 extracts consecutive intervals from the time intervals not containing noise. As described above, a consecutive interval is an interval consisting of one or more consecutive time intervals determined not to contain noise. The noise determination unit 4 may use all consecutive intervals to perform the processes of steps S1 to S4 shown in embodiments 1 and 2. However, embodiment 3 illustrates an example in which only a portion of all consecutive intervals is used to perform the processes of steps S1 to S4 shown in embodiments 1 and 2. In this case, the noise determination unit 4 performs the processes of steps S1 to S4 shown in embodiments 1 and 2 using consecutive intervals whose number of time intervals is equal to or greater than the fourth threshold value L2.

[0090] As described above, the algorithm of the noise determination unit 4 according to the first or second embodiment determines whether noise is present using the difference between the sum of spectral intensities between a time interval containing noise and a time interval containing no noise. However, if noise is present continuously throughout the entire time interval of the acquired sound data, it may be erroneously determined that noise is not present. In contrast, the abnormality diagnosis device 100 according to the third embodiment performs the noise determination according to the first or second embodiment using consecutive time intervals in which the signal E′, which is the absolute value of each element of the digital signal E of the waveform data converted by the signal converter 3, is less than the third threshold value L1, and the consecutive time intervals are equal to or greater than the fourth threshold value L2. This makes it possible to remove noise even when noise is present continuously throughout the entire time interval of the sound data. This reduces erroneous abnormality determinations due to the influence of noise.

[0091] Fourth Embodiment Fig. 14 is a block diagram showing an example of the configuration of an abnormality diagnosis device according to a fourth embodiment. As shown in Fig. 14, in an abnormality diagnosis device 100A according to the fourth embodiment, a soundproof wall 11 for reducing ambient noise is added to the configuration of the abnormality diagnosis device 100 according to the first embodiment shown in Fig. 1. The soundproof wall 11 is arranged so as to surround the object to be determined 1 and the microphone 2. The other configuration is the same as or equivalent to that in Fig. 1, and the same or equivalent components are designated by the same reference numerals, and redundant description will be omitted.

[0092] 14 , the soundproof wall 11 can be made of a general sound-absorbing material, a metal plate, or a combination thereof. It is desirable to change the thickness of the soundproof wall 11 depending on the level of ambient noise. The soundproof wall 11 may have a cylindrical shape with an open upper end and be disposed so as to surround the object to be determined 1 and the microphone 2, or may have a box-like shape with a closed upper end and house the object to be determined 1 and the microphone 2 inside.

[0093] In the fourth embodiment, the algorithm of the noise determination unit 4 is basically the same as that in the first embodiment. However, in the configuration of the first embodiment, there is a possibility that ambient noise cannot be completely removed. Also, in the configuration of the second embodiment, ambient noise cannot be completely removed, or in an environment where noise is constantly mixed in, there is a possibility that the influence of noise will remain on the abnormality determination. In the configuration of the third embodiment, as in the second embodiment, there is a possibility that all noise cannot be removed. In contrast, in the fourth embodiment, by providing the soundproof wall 11, ambient noise can be attenuated by the soundproof wall 11. Although the method of the fourth embodiment is a simple method, it is possible to provide a determination method that reduces ambient noise.

[0094] As described above, the abnormality diagnosis device 100A according to the fourth embodiment includes the soundproof wall 11 arranged to surround the object to be determined 1. The soundproof wall 11 attenuates noise from the surroundings, and therefore provides a determination method that reduces the influence of noise from the surroundings. This reduces the amount of noise mixed in.

[0095] In the above explanation, an example has been described in which the soundproof wall 11 of embodiment 4 is applied to the configuration of embodiment 1, but it goes without saying that this is not the only case, and the soundproof wall 11 of embodiment 4 can be applied to any of embodiments 1 to 4, or any of their modified examples.

[0096] The configurations shown in the above embodiments are merely examples, and may be combined with other known technologies, or different embodiments may be combined with each other. It is also possible to omit or modify parts of the configurations as long as they do not deviate from the gist of the invention.

[0097] REFERENCE SIGNS LIST 1 Object to be judged, 2 Microphone (sound collector), 3 Signal converter, 4 Noise judgment unit, 5 Abnormality judgment unit, 6 Storage unit, 10 Signal processing device, 11 Soundproof wall, 100, 100A Abnormality diagnosis device.

Claims

1. A microphone that picks up the sound generated by an object to be judged and converts it into an analog electrical signal, A signal converter that converts the analog electrical signal into a digital signal, A signal processing device that performs signal processing on the digital signal, Comprising, The signal processing device, Based on the digital signal output from the signal converter, calculates a feature amount for each state in which the time interval of the digital signal is shifted by a predetermined shift amount, and uses the feature amount to perform noise determination for determining the presence or absence of noise mixing for each time interval. A noise determination unit that performs noise determination, An abnormality determination unit that performs abnormality determination for determining the presence or absence of an abnormality in the sound of the determination object using the digital signal in the time interval where no noise is mixed output from the noise determination unit, Having, The feature amount used for the noise determination is a value indicating the spectral intensity for each time interval in the waveform data of the digital signal, An abnormality diagnosis device characterized by this.

2. The noise determination unit, Using a time-frequency analysis method for the digital signal output from the signal converter, calculates a matrix having dimensions of frequency and time, For each time interval divided by a time width including at least one time frame in the matrix, calculates the sum of the spectral intensities of one or more frequency frames included in the matrix as the feature amount, Performs the noise determination for determining the presence or absence of noise mixing for each time interval using the feature amount, The abnormality diagnosis device according to claim 1, characterized by this.

3. The time-frequency analysis method is a short-time fast Fourier transform, The abnormality diagnosis device according to claim 1 or claim 2, characterized by this.

4. The noise determination unit, Performs a short-time fast Fourier transform on the digital signal output from the signal converter to calculate a matrix having dimensions of frequency and time, For each time interval divided by a time width including at least one time frame in the matrix, calculates the sum of the spectral intensities of one or more frequency frames included in the matrix as the feature amount, Based on the feature amount, discriminates between the time interval in which noise is mixed and the time interval in which no noise is mixed in the waveform data of the digital signal output from the signal converter, Outputs the time interval in which no noise is mixed to the abnormality determination unit, The abnormality diagnosis device according to claim 1 or claim 2, characterized in that

5. The noise determination unit For each of the time intervals, compares the sum of the spectral intensities calculated as the feature amount with a preset first threshold value, Determines a time interval in which the sum of the spectral intensities is equal to or greater than the first threshold value as a time interval in which noise is mixed, Determines a time interval in which the sum of the spectral intensities is less than the first threshold value as a time interval in which no noise is mixed, The abnormality diagnosis device according to claim 3, characterized in that

6. The noise determination unit For each of the time intervals, compares the sum of the spectral intensities calculated as the feature amount with a preset first threshold value, Determines a time interval in which the sum of the spectral intensities is equal to or greater than the first threshold value as a time interval in which noise is mixed, Determines a time interval in which the sum of the spectral intensities is less than the first threshold value as a time interval in which no noise is mixed, The abnormality diagnosis device according to claim 4, characterized in that

7. The noise determination unit Performs a short-time fast Fourier transform on the digital signal output from the signal converter to calculate a matrix having two dimensions of frequency and time, For each time interval obtained by dividing the time frame in the matrix by a time width including at least one time frame, calculates the sum of the spectral intensities of one or more frequency frames included in the matrix as the feature amount, Based on the feature amount, discriminates between a time interval in which noise is mixed and a time interval in which no noise is mixed in the waveform data of the digital signal output from the signal converter, and repeats the discrimination a plurality of times by changing the time width of the time interval, Based on the discrimination results repeated a plurality of times, determines a time interval including a time frame in which the number of times discriminated as having noise mixed is equal to or greater than a preset second threshold value as a time interval in which noise is mixed, Outputs to the abnormality determination unit a time interval obtained by excluding the determined time interval in which noise is mixed as a time interval in which no noise is mixed, The abnormality diagnosis device according to claim 1 or claim 2, characterized in that

8. The noise determination unit performs pre-processing before performing the noise determination, The pre-processing Calculates the absolute value of the digital signal output from the signal converter, A process of determining, as a time period during which noise is mixed in, a time period in which the absolute value is equal to or greater than a preset third threshold value. The noise determination unit performs the noise determination using a time period obtained by excluding the time period during which the noise determined in the previous-stage process is mixed in. The abnormality diagnosis apparatus according to claim 1 or claim 2, characterized in that.

9. The noise determination unit performs the noise determination using a continuous period in which the time period without the mixed noise, which is obtained by excluding the time period during which the noise determined in the previous-stage process is mixed in, is continuous for a preset fourth threshold value or more. The abnormality diagnosis apparatus according to claim 8, characterized in that.

10. A soundproof wall arranged to cover the surroundings of the object to be determined and the sound collector The abnormality diagnosis apparatus according to claim 1 or claim 2, characterized by comprising.

11. A step of collecting a sound generated by an object to be determined and converting it into an analog electrical signal; A step of converting the analog electrical signal into a digital signal; Based on the digital signal, calculating a feature amount for each time period of the digital signal, and performing a noise determination to determine the presence or absence of mixed noise for each state in which the time period is shifted by a predetermined shift amount using the feature amount; A step of performing an abnormality determination to determine the presence or absence of an abnormality in the sound of the object to be determined using the digital signal of the time period determined to have no mixed noise by the noise determination; Comprising The feature amount used for the noise determination is a value indicating the spectral intensity for each time period in the waveform data of the digital signal. An abnormality diagnosis method, characterized in that.

12. A step of collecting a sound generated by an electrical device, which is an object to be determined, and converting it into an analog electrical signal; A step of converting the analog electrical signal into a digital signal; Based on the digital signal, calculating a feature amount for each time period of the digital signal, and performing a noise determination to determine the presence or absence of mixed noise for each state in which the time period is shifted by a predetermined shift amount using the feature amount; A step of performing an abnormality determination to determine the presence or absence of an abnormality in the sound of the object to be determined using the digital signal of the time period determined to have no mixed noise by the noise determination; Comprising The feature amount used for the noise determination is a value indicating the spectral intensity for each of the time intervals in the waveform data of the digital signal, determining the electric device determined to be normal by the abnormality determination as a non-defective product, A method for manufacturing an electric device, characterized by the above.