JP7253721B2 - Abnormal noise determination device, abnormal noise determination method, and abnormal noise determination system - Google Patents

Description

The present disclosure relates to an abnormal noise determination device, an abnormal noise determination method, and an abnormal noise determination system that determine abnormal noise emitted from a target device.

If equipment that is always in operation during business hours, such as facility equipment in a building or equipment in a factory, stops due to an abnormality, production will not be possible, resulting in serious damage such as a decrease in production efficiency or the occurrence of harm. may occur. In order to detect such abnormalities in equipment in advance, it is considered effective to monitor abnormalities using sensors capable of detecting current, vibration, sound, and the like.

For example, Japanese Unexamined Patent Application Publication No. 2002-100000 discloses an abnormal noise determination device that determines abnormal noise emitted from a target device such as a steel tower or a power transmission line. Disturbance noise generated by wind noise and the like is not usually an abnormal noise. This abnormal noise determination device uses two types of sample data, a plurality of sound data including abnormal sounds and a plurality of sound data not including abnormal sounds, in order to efficiently determine abnormal sounds in the presence of disturbance sounds. to determine the presence or absence of abnormal noise.

JP 2014-145645 A

However, in the abnormal noise determination apparatus of Patent Document 1, when disturbance noise including various sounds that occur irregularly has a frequency spectrum similar to the abnormal sound emitted from the device to be inspected, the device to be inspected It becomes difficult to identify whether or not the sound emitted from the device contains an abnormal sound.

Further, in order to identify the presence or absence of abnormal sounds, it takes a lot of time and effort to collect a plurality of sound data containing abnormal sounds and a plurality of sound data not containing abnormal sounds as sample data. In addition, the amount of sample data that is sufficient to ensure the accuracy of identifying the presence or absence of abnormal noise varies greatly depending on the type of equipment to be inspected or its installation environment, so it is difficult to determine the appropriate amount of sample data. be.

The present disclosure has been devised in view of the conventional situation described above. It is an object of the present invention to provide an abnormal noise determination device, an abnormal noise determination method, and an abnormal noise determination system that can improve the accuracy of identifying whether or not there is an abnormality and appropriately detect the presence or absence of an abnormality in a target device.

The present disclosure includes an input unit connected to a microphone that picks up sound emitted from a target device, a memory that stores sound data for a predetermined period of the sound emitted from the target device, and sound data for the predetermined period. and a processor that determines whether the sound corresponding to the sound data in the one frame period is a stationary sound or a sudden sound for each one frame period shorter than the predetermined period, and the memory is stored in the learning mode during the learning mode. Stores at least one stationary sound as a normal sound from among the one or more of the one-frame-period sound data determined to be the stationary sound, and the processor stores the stationary sound as a normal sound during an operation mode different from the learning mode. identifying whether or not the sound emitted from the target device includes an abnormal sound in accordance with a comparison between the determined sound data of the one frame period and the sound data of the normal sound stored in the memory; An abnormal noise determination device is provided.

Further, the present disclosure is an abnormal noise determination method executed by an abnormality determination device, wherein sound data for a predetermined period of sound emitted from the target device is input from a microphone that picks up the sound emitted from the target device. and determining whether the sound corresponding to the sound data for the one-frame period is a stationary sound or a sudden sound for each one-frame period shorter than the predetermined period, using the input sound data for the predetermined period. a step of obtaining at least one stationary sound stored as a normal sound from among the one or more of the one-frame-period sound data determined to be the stationary sound during the learning mode; Whether or not abnormal sound is included in the sound emitted from the target device in accordance with a comparison between the sound data of the one-frame period determined to be the stationary sound and the sound data of the normal sound acquired during the operation mode. and identifying whether the noise is abnormal.

Further, the present disclosure includes an input unit connected to a microphone that collects sound emitted from a target device, a memory that stores sound data for a predetermined period of sound emitted from the target device, and a sound data for the predetermined period. a processor that uses sound data and determines, for each frame period shorter than the predetermined period, whether the sound corresponding to the sound data in the one frame period is a stationary sound or a sudden sound, wherein the memory is in a learning mode. store as a normal sound at least one stationary sound from among the sound data of one or more frame periods determined as the stationary sound during the operation mode, and the processor stores the stationary sound during the operation mode different from the learning mode. Identifying whether or not the sound emitted from the target device includes an abnormal sound in accordance with a comparison between the sound data of the one frame period determined to be a sound and the sound data of the normal sound stored in the memory. To provide an abnormal noise determination system.

According to the present disclosure, even if the sound collected from the target device includes not only the stationary sound but also the disturbance sound such as a sudden sound, the accuracy of identifying whether or not the collected sound includes an abnormal sound is improved. It is possible to appropriately detect the presence or absence of an abnormality in the target device.

1 is a diagram showing an example of a schematic configuration of an abnormal noise determination system according to Embodiment 1. FIG. Diagram showing the hardware configuration of the abnormal noise determination system Flowchart showing sound data learning operation procedure Flowchart showing steady sound extraction procedure in step S4 Flowchart showing the minimum stationary detection procedure in step S5 Characteristic diagram showing temporal changes in sound pressure level and frequency distribution of sound picked up in learning mode Flowchart showing operation procedure of sound data Flowchart showing normal/abnormal identification procedure in step S16 Characteristic diagrams showing temporal changes in the sound pressure level and frequency distribution of sound in normal and abnormal states picked up in operation mode. A diagram showing an example of a schematic configuration of an abnormal noise determination system according to a second embodiment. Diagram showing the hardware configuration of the abnormal noise determination system Flowchart showing sound data learning operation procedure Flowchart showing operation procedure of sound data Flowchart showing normal/abnormal identification procedure in step S65

Hereinafter, embodiments specifically disclosing an abnormal noise determination device, an abnormal noise determination method, and an abnormal noise determination system according to the present disclosure will be described in detail with reference to the drawings as appropriate. However, more detailed description than necessary may be omitted. For example, detailed descriptions of well-known matters and redundant descriptions of substantially the same configurations may be omitted. This is to avoid unnecessary verbosity in the following description and to facilitate understanding by those skilled in the art. It should be noted that the accompanying drawings and the following description are provided for a thorough understanding of the present disclosure by those skilled in the art and are not intended to limit the claimed subject matter.

(Embodiment 1)
FIG. 1 is a diagram showing an example of a schematic configuration of an abnormal noise determination system 5 according to Embodiment 1. As shown in FIG. The abnormal sound determination system 5 includes a microphone 10 , an AD converter 20 and an information processing device 30 . The microphone 10 picks up sound emitted from the device 100 to be inspected regularly for purposes such as monitoring. Devices to be inspected include, for example, servers installed in companies or data centers, and compressors and motors installed in factories. When abnormal noise is frequently emitted from the target device 100 to be inspected, the information processing device 30 determines that the target device 100 is in an abnormal state.

The microphone 10 is installed, for example, in a machine room where the device 100 to be inspected is installed. The microphone 10 has its sound receiving surface facing the target device 100, picks up the sound emitted by the target device 100, converts the input sound wave into an electric signal, and outputs the electric signal as a sound signal. The number of microphones 10 may be one or plural. For example, the microphone may be a single high-quality electret condenser microphone (ECM) or a microphone array composed of a plurality of ECMs.

The AD converter 20 is installed in the same machine room as the microphone 10 . The AD converter 20 receives sounds picked up by one or a plurality of microphones 10, analog-to-digital converts analog sound signals simultaneously or in a time-division manner, and outputs digital sound data. The AD converter 20 converts an analog signal into a digital signal with a predetermined quantization bit and sampling frequency. The number of AD converters 20 may be the same as the number of microphones 10 or may be less than the number of microphones 10 .

The information processing device 30 is a monitoring device that constantly monitors the state of the inspection target device 100 . The information processing device 30 is configured by, for example, a PC (Personal Computer) or a tablet terminal. The information processing device 30 is installed in a monitoring room different from the machine room in which the target device is installed.

The information processing device 30 is placed in a monitoring room that is separate from the machine room. All of the microphone 10, the AD converter 20 and the information processing device 30 may be arranged in the same machine room or monitoring room.

FIG. 2 is a diagram showing the hardware configuration of the abnormal noise determination system 5. As shown in FIG. The information processing device 30 is connected to the AD converter 20 and receives digital sound data output from the AD converter 20 . Based on the input sound data, the information processing apparatus 30 determines whether or not the sound emitted by the inspection target device 100 includes an abnormal sound (abnormal sound determination processing).

In this abnormal sound determination process, the information processing device 30 performs machine learning such as deep learning in advance to generate a learned model. Note that the information processing apparatus 30 may copy and use a trained model generated by another computer. Here, the trained model is generated as a program for executing a steady sound extraction process (see FIG. 4) for extracting a steady sound frame by frame from sound data for a predetermined time. Also, the learned model is generated as a program for executing the normal/abnormal identification process (see FIG. 8) for determining whether or not the sound emitted from the target device includes an abnormal sound. The information processing device 30 uses the generated trained model to perform steady sound extraction processing and normal/abnormal identification processing. Note that the information processing device 30 may perform the steady sound extraction process and the normal/abnormal identification process using a cross-correlation function without using a machine-learned model. Further, the information processing device 30 may be configured to execute processing by a device such as one computer, or may be configured to execute processing by devices such as a plurality of computers. is not limited to

Information processing apparatus 30 includes processor 301 , memory 302 , storage 303 , communication section 304 , operation section 305 , display 306 , and input section 307 .

The processor 301 is configured using various processing devices such as a CPU (Central Processing Unit), a DSP (Digital Signal Processor), and an FPGA (Field Programmable Gate Array), and comprehensively executes processing related to sound data.

The memory 302 has a memory device such as a RAM (Random Access Memory), is used as a working memory of the processor 301, and is used for temporary storage in calculations during data processing. The memory 302 has a memory device such as a ROM (Read Only Memory), and stores various execution programs for executing the processing of the processor 301 and various setting data related to processing such as machine learning.

The storage 303 is configured using various storage devices such as HDD (Hard Disk Drive), SSD (Solid State Drive), and optical disk drive, and stores data such as target sound data and learning models generated by machine learning. .

A communication unit 304 is an interface that performs wired or wireless communication.

The operation unit 305 receives user operations, and may include a mouse, keyboard, touch pad, touch panel, microphone, or other input device.

The display 306 displays an abnormal sound detection notification generated by the processor 301 as a result of the determination that the abnormal sound is included. Display 306 may include a liquid crystal display device, organic EL device, or other display device. Note that the display may be a remote monitoring monitor that can be controlled by the information processing device 30 . Also, the operation unit 305 and the display 306 may be an integrated touch panel.

The input unit 307 inputs digital sound data picked up by the microphone 10 and output from the AD converter 20 .

Next, the operation of the abnormal noise determination system 5 according to Embodiment 1 will be described.

The abnormal noise determination system 5 operates in a learning mode and an operation mode to perform abnormal noise determination processing. The learning mode is an operation mode in which the smallest stationary sound among the stationary sounds contained in the sound emitted from the target device is learned and detected, and the detection result is stored in the information processing device 30 as the learning result. On the other hand, the operation mode is an operation mode in which, in an actual operating environment other than the learning mode, it is determined whether or not the sound emitted from the target device contains abnormal sounds using the learning results stored in the learning mode. be.

(learning mode)
In the learning mode, the processor 301 extracts a plurality of stationary sounds included in the sounds picked up by the microphone 10 and stores sound data of the minimum stationary sound among the plurality of extracted stationary sounds in the storage 303 . The plurality of steady sounds include sounds emitted when the device 100 to be tested is in a normal state. The learning may be performed only once in the initial state in which the target device 100 to be inspected is installed, or may be performed periodically during maintenance or the like.

FIG. 3 is a flow chart showing a learning operation procedure for sound data. The microphone 10 mainly picks up sounds emitted from the device 100 to be tested. The processor 301 inputs sound data picked up by the microphone 10 and analog-to-digital converted by the AD converter 20 (S1). The processor 301 accumulates the input sound data in the storage 303 . The processor 301 determines whether or not a predetermined period of time has elapsed and recording by the microphone 10 has ended (S2). Recording by the microphone 10 is performed for a predetermined time. The predetermined time is any fixed time such as 30 minutes, 1 hour, or 1 day. Also, the predetermined time may be the time until the user (including the administrator) manually stops, instead of the fixed time. Further, it is preferable to set the predetermined time to the recording time (fixed time) in the operation mode in order to perform the abnormal sound determination process.

If the recording is not finished in step S2, the process of processor 301 returns to step S1. On the other hand, when recording ends in step S2, the processor 301 performs frequency analysis on the time-series sound data accumulated in the storage 303 (S3). This frequency analysis is performed, for example, by FFT (Fast Fourier Transform) on sound data in units of 100 ms. Based on the result of frequency analysis, the processor 301 performs stationary sound extraction processing for extracting stationary sounds contained in the sound data on a frame-by-frame basis (S4). A frame (one frame) is, for example, a sound length for determining whether sound data is a stationary sound or a sudden sound. The length of one frame varies depending on the type of equipment to be inspected, and is arbitrarily set within a range of 10 ms to 500 ms, for example. Also, one frame is set wider than the time width of the sudden sound (for example, the pulse width in the case of a pulse-like sudden sound). The process of extracting this stationary sound is performed using a machine-learned model. Note that this process may be performed using a cross-correlation function.

The processor 301 accumulates the sound data of one or more extracted stationary sounds in the storage 303 . The processor 301 performs minimum stationary sound detection processing for detecting a stationary sound having the lowest spectral power (hereinafter referred to as minimum stationary sound) from one or more stationary sounds (S5). The processor 301 stores this minimum steady sound in the storage 303 as a steady sound (normal sound) when the device 100 to be inspected is in a normal state (S6). Note that the processor 301 may store sound data of the minimum stationary sound in an external storage connected via the communication unit 304 . After this, the processor 301 terminates the operation in the learning mode.

FIG. 4 is a flow chart showing the steady sound extraction procedure in step S4. The processor 301 reads sound data for a certain period of time from the sound data accumulated in the storage 303 . The fixed time may be, for example, 10 seconds, 30 minutes, or a time corresponding to a predetermined number of frames (20 frames or 50 frames). The processor 301 accumulates the spectral power of sound data for a certain period of time, and divides this accumulated value by one frame period to calculate the average spectral power per frame (S21).

The processor 301 sequentially determines whether or not the spectral power of the sound data for one frame period exceeds the average spectral power among the sound data for a certain period of time (S22). The spectral power of sound data in one frame period is obtained by calculating and averaging the acoustic power of each frequency for each frame. If the spectral power of the sound data for one frame period is less than or equal to the average spectral power, the processor 301 determines that the sound data for one frame period is stationary sound (S23).

The processor 301 determines whether or not the one-frame sound data determined to be stationary sound has frequency characteristics similar to the stationary sound data already stored in the storage 303 (S24). Judgment of the similarity of the sound data is performed using a trained model by machine learning. Note that this similarity determination may be made using a cross-correlation function.

If it is determined that one frame of sound data is similar to the sound data of the stationary sound stored in the storage 303, the processor 301 proceeds to the process of step S26. On the other hand, if it is determined in step S24 that one frame of sound data is not similar to the sound data of the stationary sound stored in the storage 303, the processor 301 regards this one frame of sound data as a new stationary sound. Store in the storage 303 (S25).

After the processing of steps S24 and S25, the processor 301 determines whether or not all sound data in units of frames for a certain period of time have been processed (S27). For example, if one frame period is 0.5 seconds (500 ms) and the fixed time is 30 minutes, the processor 301 generates a stationary sound for 3600 (=60/0.5×30) frames of sound data. Extraction processing will be performed. If all the frame-by-frame sound data for the predetermined period of time have not been processed, the process of the processor 301 returns to step S22. On the other hand, when all the frame-by-frame sound data for the predetermined period of time have been processed in step S27, the processor 301 terminates the steady sound extraction process and returns to the main process.

If the spectral power of the sound data for one frame period exceeds the average spectral power in step S22, the processor 301 determines that this one frame of sound data is a sudden sound (S26). Processing of the processor 301 proceeds to step S27.

Even if one frame of sound data includes a sudden sound, if the power of one sudden sound is small and the spectral power is equal to or less than the average spectral power in step S22, the processor 301 determines that the sound is a stationary sound. . On the other hand, even if the power of one sudden sound is small, if the sound data of one frame contains many sudden sounds and the spectral power exceeds the average spectral power, the processor 301 determines that it is a sudden sound.

FIG. 5 is a flow chart showing the minimum stationary sound detection procedure in step S5. The processor 301 reads sound data of stationary sound for a certain period of time accumulated in the storage 303 . The processor 301 mutually compares the spectral powers of one or more stationary sounds (S31). As a result of the comparison, the processor 301 stores the stationary sound with the lowest spectral power in the storage 303 as the minimum stationary sound (S32). After that, the processor 301 ends this process and returns to the main process.

In this way, the processor 301 compares the spectral powers of one or more stationary sounds and selects the minimum stationary sound from among the one or more stationary sounds so as not to select stationary sounds containing disturbance sounds. to Therefore, the processor 301 can exclude stationary sounds including disturbance sounds from comparison targets in the normal/abnormal identification process in the operation mode.

FIG. 6 is a characteristic diagram showing temporal changes in sound pressure level and frequency distribution of sound picked up in the learning mode. A characteristic diagram gh1 represents the time change of the sound pressure level. A characteristic diagram gh2 represents the time change of the frequency distribution. The horizontal axes of the characteristic diagrams gh1 and gh2 represent time. The vertical axis of the characteristic diagram gh1 represents the sound pressure level. The vertical axis of the characteristic diagram gh2 represents frequency. Also, the characteristic diagram gh2 expresses the sound pressure level by color changes. Here, regions with high sound pressure levels are represented in yellow (dark dots in the figure). A region with an intermediate sound pressure level is represented in red (intermediate dots in the figure). A region with a low sound pressure level is indicated by magenta (light dots in the figure).

In the learning mode, sound data is collected for a certain period of time (for example, 30 minutes) and accumulated in the storage 303 . Characteristic diagrams gh1 and gh2 show sound data for 10 seconds in 30 minutes. Here, it is assumed that the interval for judging the state of sound data is one frame, and one frame is set to 0.5 seconds (500 msec). One frame can be arbitrarily set, for example, from 10 ms to 500 ms. If one frame is 500 msec, 10 seconds of sound data is represented by 20 frames of sound data.

In the first frame F1 (0 to 500 ms), stationary sound data with a low sound pressure level continues. Therefore, the first frame F1 is a section of stationary sound in which the spectral power of the sound is extremely small. The second frame F2 (500-1000 ms) is in the same situation as the first frame F1 (0-500 ms).

In the third frame F3 (1000 to 1500 ms), a sudden sound with a high sound pressure level is generated just after 1000 m. Therefore, the third frame F3 is a section of a sudden sound in which the spectral power of the sound is large. The fourth frame F4 (1500-2000 ms) is in the same situation as the first frame F1 (0-500 ms).

In the fifth frame F5 (2000 to 2500 ms), continuous sound data with a slightly higher sound pressure level is generated. Such continuous sound data includes, for example, motor sound. The fifth frame F5 is a section of stationary sound in which the spectral power of the continuous sound data is not so high and the spectral power is slightly high. In the sixth frame F6 (2500 to 3000 ms), following the fifth frame F5, continuous sound data with a slightly higher sound pressure level is generated. A high-level sudden sound is occurring. Therefore, the sixth frame F6 is a section of a sudden sound in which the spectral power of the sound is large. The seventh frame F7 (3000-3500 ms) and the eighth frame F8 (3500-4000 ms) are in the same situation as the fifth frame F5 (2000-2500 ms).

In the ninth frame F9 (4000 to 4500 ms), there is no continuous sound data with a slightly high sound pressure level, and steady sound data with a low sound pressure level continues as in the first frame F1 (0 to 500 ms). ing. Therefore, the ninth frame F9 is a section of stationary sound in which the spectral power of the sound is extremely small. The tenth frame F10 (4500-5000 ms) is in the same situation as the ninth frame F9.

In the eleventh frame F11 (5000 to 5500 ms), sudden sounds with extremely high sound pressure levels occur frequently. Therefore, the 11th frame F11 is a section of a sudden sound in which the spectral power of the sound is extremely large. In the twelfth frame F12 (5500 to 6000 ms), a sudden sound with a slightly high sound pressure level is generated. The twelfth frame F12 is determined to be a sudden sound section in which the spectral power of the sound exceeds the average spectral power. The average spectral power is, for example, an average value of spectral powers of sounds calculated within a certain period of time (30 minutes) or 20 frames (10 seconds).

The 13th frame F13 (6000-6500 ms) and the 14th frame F14 (6500-7000 ms) are in the same situation as the 9th frame F9.

The 15th frame F15 (7000 to 7500 ms) and the 16th frame F16 (7500 to 8000 ms) generate continuous sound data with high sound pressure levels. The fifth frame F5 is a section of a large non-stationary sound in which the spectral power of continuous sound data is large and the spectral power exceeds the average spectral power.

The 17th frame (8000 to 8500 ms) to the 20th frame (9500 to 10000 ms) are in the same situation as the 14th frame F14.

As described above, when the spectral power of sound does not exceed the average spectral power in each frame, the sound data is determined to be a stationary sound even if a slightly high-pitched sound is continuously generated. On the other hand, if the spectral power of the sound exceeds the average power in each frame, the sound data is determined to be a sudden sound. The processor 301 accumulates in the storage 303 all of the plurality of sound data determined to be stationary sounds, and is used when acquiring sound data in a normal state.

(operation mode)
In the operational mode, the processor 301 detects the minimum stationary sound at regular time intervals (for example, 30 minutes, 1 day) in the same manner as in the learning mode. The processor 301 compares the current minimum stationary sound with the minimum stationary sound stored in the storage 303 when the inspection target device 100 is in a normal state. The processor 301 determines the normal state/abnormal state of the target device 100 based on this comparison result.

FIG. 7 is a flow chart showing a procedure for operating sound data. Operational operations are performed regularly or irregularly. In the operation mode, the microphone 10 picks up sounds emitted from the device 100 to be tested. The processor 301 inputs sound data picked up by the microphone 10 and analog-to-digital converted by the AD converter 20 (S11). The processor 301 accumulates the input sound data in the storage 303 .

The processor 301 determines whether or not a certain period of time has passed since the sound pickup by the microphone 10 was started (S12). The certain period of time is set to a longer period of time in order to increase the accuracy of extracting stationary sounds, and is an arbitrary certain period of time such as 30 minutes, one hour, or one day. Further, it is preferable that the predetermined period of time is equal to the predetermined period of time in the learning mode in order to perform the abnormal noise determination process. If the fixed time has not passed, the processor 301 returns to the process of step S11. Although a case where the processor 301 automatically starts the operation (for example, by starting a timer) is shown here, the user may manually instruct the start of the operation at any time. When a user (including an administrator) gives an instruction to start operation, the processor 301 may accept this instruction via the operation unit 305 .

When the predetermined time has passed in step S12, the processor 301 performs frequency analysis on the time-series sound data accumulated in the storage 303 (S13). This frequency analysis is the same as in the learning mode.

Based on the result of the frequency analysis, the processor 301 performs stationary sound extraction processing for extracting stationary sounds in units of frames included in the sound data (S14). This stationary sound extraction processing is the same as in the learning mode. The processor 301 accumulates the sound data of one or more extracted stationary sounds in the storage 303 . The processor 301 performs minimum stationary sound detection processing for detecting the minimum stationary sound from among one or more stationary sounds (S15). This minimum stationary sound detection processing is the same as in the learning mode.

Based on the minimum steady sound detected in step S15, the processor 301 performs normal/abnormal identification processing for identifying whether the device 100 to be inspected is in a normal state or an abnormal state (S16). In this normal/abnormal identification process, by comparing the minimum stationary sounds, stationary sounds including continuously occurring disturbance sounds and infrequently occurring sudden sounds are excluded from comparison targets. Therefore, it is possible to improve the accuracy of normal/abnormal discrimination. The details of this normal/abnormal identification process will be described later.

The processor 301 determines whether or not to end the operational operation (S17). The end of the operational operation is recognized when the user (including the administrator) manually instructs to stop and the processor 301 accepts this instruction via the operation unit 305 . Note that the operation operation may be terminated automatically by the processor 301 (for example, timer termination). In this case, manual operation by the user can be omitted. When not ending the operation operation, the processor 301 returns to the process of step S11 and repeats the same process. Note that the processor 301 may return to the process of step S13 and repeat the same process. In this case, abnormal sound determination is repeatedly performed on the sound data accumulated in steps S11 and S12 for a certain period of time. On the other hand, when ending the operation operation in step S17, the processor 301 ends this operation operation.

FIG. 8 is a flow chart showing the normal/abnormal identification procedure in step S16. The processor 301 reads the sound data of the minimum stationary sound stored in the storage 303 as normal sound in the learning mode (S41). The processor 301 calculates the degree of similarity between the sound data of the current minimum stationary sound detected in step S15 and the sound data of the normal sound (S42). This degree of similarity calculation may be performed using a machine-learned model or a cross-correlation function.

The processor 301 determines whether or not the similarity calculated in step S42 is equal to or greater than the threshold TH1 (S43). The threshold TH1 is a value for the processor 301 to determine whether or not sound data of stationary sounds are similar. When the degree of similarity is less than the threshold TH1, the processor 301 determines that the device 100 to be inspected is emitting an abnormal sound (S44). The processor 301 performs notification processing (S45). In this notification process, the processor 301 displays an abnormality detection notification indicating an abnormality on the display 306 . Note that the processor 301 notifies a remote monitoring device (not shown) connected to the network via the communication unit 304 that the device 100 to be inspected is in an abnormal state by emitting an abnormal sound. may notify you. Thereafter, the processor 301 terminates this operational operation and returns to the main processing.

When the degree of similarity is equal to or greater than the threshold TH1 in step S43, the processor 301 determines that the target device 100 under test is emitting normal sounds (S46). In this case, the device 100 to be inspected is in a normal state, so the processor 301 does not perform notification processing. Note that when the device 100 to be inspected is in a normal state, the processor 301 may perform notification processing, such as displaying on the display 306 that it is normal.

FIG. 9 is a characteristic diagram showing temporal changes in the sound pressure level and frequency distribution of sounds in normal and abnormal states collected in the operation mode.

A characteristic diagram gh3 in the normal state represents the time change of the sound pressure level. A characteristic diagram gh4 in a normal state represents the time change of the frequency distribution. The horizontal axes of the characteristic diagrams gh3 and gh4 in the normal state represent time. The vertical axis of the normal state characteristic diagram gh3 represents the sound pressure level. The vertical axis of the normal state characteristic diagram gh4 represents the frequency. In addition, the characteristic diagram gh4 in the normal state represents the sound pressure level with a change in color. Here, regions with high sound pressure levels are represented in yellow (dark dots in the drawing). A region with an intermediate sound pressure level is represented in red (intermediate dots in the figure). A region with a low sound pressure level is indicated by magenta (light dots in the figure).

Similarly, the abnormal state characteristic diagram gh5 represents the time change of the sound pressure level. A characteristic diagram gh6 of an abnormal state represents the time change of the frequency distribution. The horizontal axes of the abnormal state characteristic diagrams gh5 and gh6 represent time. The vertical axis of the abnormal state characteristic diagram gh5 represents the sound pressure level. The vertical axis of the abnormal state characteristic diagram gh6 represents the frequency. A characteristic diagram gh6 of an abnormal state represents the sound pressure level with a change in color. Here, regions with high sound pressure levels are represented in yellow (dark dots in the drawing). A region with an intermediate sound pressure level is represented in red (intermediate dots in the figure). A region with a low sound pressure level is indicated by magenta (light dots in the figure).

In addition, in the characteristic of the sound data shown in FIG. 9, the sudden sound is generated at the same timing in the normal state and the abnormal state of the target device, but this characteristic is an example. In many cases, the timing at which the sudden sound is generated may differ between the normal state and the abnormal state of the target device.

In the operation mode, sound data is collected for a certain period of time (for example, 30 minutes) and accumulated in the storage 303 . Characteristic diagrams gh3 and gh4 and characteristic diagrams gh5 and gh6 show sound data for 10 seconds in 30 minutes. As in the learning mode, when one frame is 500 msec, 10 seconds of sound data is represented by 20 frames of sound data.

In FIG. 9, as an example, the sound data in the normal state of the target device in the operational mode is the same as the sound data shown in FIG. 6 in the learning mode. Therefore, the description of the temporal changes in the sound pressure level and frequency distribution in each frame of the sound data picked up in the normal state will be omitted. Here, the characteristics of the sound data in the abnormal state of the target device will be compared with the characteristics of the sound data in the normal state on a frame-by-frame basis.

In the first frame F31 (0 to 500 ms), steady sound data with a low sound pressure level continues in the normal state. On the other hand, in an abnormal state, an abnormal noise with a slightly high sound pressure level is intermittently generated and superimposed on steady sound data with a low sound pressure level. However, in the first frame F31, the abnormal state sound data is determined to be a stationary sound because it has spectral power equivalent to that of the surrounding frames. In other words, the first frame F31 is a section of steady sound in which the spectral power of the sound is extremely small in the normal state, and is a section of the stationary sound in which the spectral power of the sound is small even in the abnormal state.

The second frame F32 (500-1000 ms) is in almost the same situation as the first frame F31 (0-500 ms).

In the third frame F33 (1000 to 1500 ms), in the normal state, a sudden sound with a high sound pressure level is generated just after 1000 m. Further, in an abnormal state, a sudden sound with a high sound pressure level is superimposed on sound data in which an abnormal sound with a slightly high sound pressure level is intermittently generated. Therefore, the third frame F33 is a section of a sudden sound in which the spectral power of the sound is large in both the normal state and the abnormal state. The fourth frame F34 (1500-2000 ms) is in the same situation as the first frame F31.

In the fifth frame F35 (2000 to 2500 ms), continuous sound data with a slightly high sound pressure level is generated in the normal state. Such continuous sound data includes, for example, motor sound. The fifth frame F35 is a section of stationary sound in which the spectral power of the continuous sound data is not so high and the spectral power is slightly high. In addition, in an abnormal state, an abnormal sound with a slightly high sound pressure level is intermittently generated and superimposed on the continuous sound data with a slightly high sound pressure level. The fifth frame F35 is a section of steady sound in both normal and abnormal states.

In the sixth frame F36 (2500 to 3000 ms), in the normal state, following the fifth frame F35, continuous sound data with a slightly higher sound pressure level is generated. At this point, a sudden sound with a high sound pressure level is generated. In the abnormal state, an abnormal sound with a slightly high sound pressure level is intermittently generated, and a sudden sound with a high sound pressure level is superimposed on continuous sound data with a slightly high sound pressure level. Therefore, the sixth frame F36 is a section of a sudden sound in which the spectral power of the sound is large in both the normal state and the abnormal state. The seventh frame F37 (3000-3500 ms) and the eighth frame F38 (3500-4000 ms) are in the same situation as the fifth frame F35.

In the ninth frame F39 (4000 to 4500 ms), there is no continuous sound data with a slightly high sound pressure level in the normal state. data continues. In addition, in an abnormal state, an abnormal sound with a slightly high sound pressure level is intermittently generated and superimposed on steady sound data with a low sound pressure level. The ninth frame F39 is a stationary sound section in both the normal state and the abnormal state. The tenth frame F40 (4500 to 5000 ms) is in almost the same situation as the ninth frame F39.

In the eleventh frame F41 (5000 to 5500 ms), sudden sounds with extremely high sound pressure levels are frequently generated in the normal state. In addition, in an abnormal state, sudden sounds with extremely high sound pressure levels are frequently generated, and abnormal sounds with slightly high sound pressure levels are generated intermittently. Therefore, the 11th frame F41 is a section of a sudden sound in which the spectral power of the sound is extremely large in both the normal state and the abnormal state.

In the twelfth frame F42 (5500 to 6000 ms), a sudden sound with a slightly high sound pressure level is generated in the normal state. In addition, in the abnormal state, a sudden sound with a slightly high sound pressure level is generated, and an abnormal sound with a slightly high sound pressure level is generated intermittently. Therefore, the twelfth frame F42 is determined to be a sudden sound section in which the spectral power of the sound exceeds the average spectral power in both the normal state and the abnormal state.

The 13th frame F43 (6000-6500 ms) and the 14th frame F44 (6500-7000 ms) are in the same situation as the 9th frame F9.

In the 15th frame F45 (7000-7500 ms) and the 16th frame F46 (7500-8000 ms), continuous sound data with a high sound pressure level are generated in the normal state. In addition, in an abnormal state, continuous sound data with a high sound pressure level is generated, and abnormal noise with a slightly high sound pressure level is generated intermittently. The 15th frame F45 and the 16th frame F46 are sections of non-stationary sound in which the spectral power of continuous sound data exceeds the average spectral power in both the normal state and the abnormal state.

The 17th frame F47 (8000-8500 ms), the 18th frame F48 (8500-9000 ms) and the 19th frame F49 (9000-9500 ms) are in the same situation as the 13th frame F43 and 14th frame F44.

In the twentieth frame F50 (9500 to 10000 ms), steady sound data with a low sound pressure level continues in both the normal state and the abnormal state. Also, no abnormal noise is generated in the abnormal state. The twentieth frame F50 is a section of stationary sound in which the spectral power of the sound is extremely small in both the normal state and the abnormal state.

Here, when the processor 301 performs normal/abnormal identification processing using the minimum steady sound of the 20th frame F50, even if the target device is in an abnormal state, the current minimum steady sound and the normal sound stored in the learning mode becomes equal to or greater than the threshold value TH1, and it is determined that the state is normal. In this case, it is desirable that the processor 301 performs the normal/abnormal identification processing using the minimum stationary sound of the frames excluding the twentieth frame F50. Also, even if there is sound data in the twentieth frame F50, the abnormal sound determination process is repeatedly executed by the processor 301 using the sound data accumulated in the storage 303 for a certain period of time. Judgment is possible.

As described above, in the abnormal sound determination system 5 according to Embodiment 1, the processor 301 extracts the stationary sound from the sound data of the sound emitted by the target device 100, and compares it with the stationary sound in the normal state stored in advance. Thus, normal/abnormal identification processing of the target device 100 is performed. By comparing the stationary sounds, the processor 301 can improve the accuracy of normal/abnormal identification even in an environment where the sounds picked up by the microphone include disturbance sounds. In particular, the processor 301 compares the minimum steady sound in the normal state with the minimum steady sound at the present time, so that the sound data emitted by the target device 100 includes disturbance sounds (for example, sudden sounds and continuous sounds with a slightly high sound pressure level). The abnormal state of the target device 100 can be properly detected even when the target device 100 includes a motor sound or the like. In addition, the processor 301 learns the minimum stationary sound at the initial stage of startup, such as when the target device 100 is installed, so that the target device 100 can learn the normal sound in a favorable state. Moreover, the abnormal noise determination system 5 can cope with the installation of the target device 100 in various environments. Note that the abnormal sound determination system is not limited to the configuration including one microphone and one AD converter, and may be configured to include an arbitrary number of microphones and an arbitrary number of AD converters.

As described above, the information processing device 30 (an example of the abnormal sound determination device) includes the input unit 307 connected to the microphone 10 that picks up the sound emitted from the target device 100, and the sound emitted from the target device 100 for a predetermined period. Using a storage 303 (an example of a memory) for storing sound data and sound data for a predetermined period, the sound corresponding to the sound data for one frame period shorter than the predetermined period is a stationary sound or a sudden sound. and a processor 301 for determining whether there is The storage 303 stores, as a normal sound, at least one stationary sound among the sound data of one or more frame periods determined as stationary sounds during the learning mode. The processor 301 compares the sound data of one frame period determined to be a steady sound during the operation mode different from the learning mode with the sound data of the normal sound stored in the storage 303 . It is determined (an example of identification) whether or not the sound contains an abnormal sound.

As a result, even if the sound picked up from the target device 100 includes not only a stationary sound but also a disturbance sound such as a sudden sound, the information processing apparatus 30 can determine whether the picked-up sound includes an abnormal sound. It is possible to improve the identification accuracy of the target device and appropriately detect the presence or absence of anomalies in the target device.

Further, when the processor 301 identifies that the sound emitted from the target device 100 includes an abnormal sound, the processor 301 displays on the display 306 an abnormal sound detection notification indicating that the sound emitted from the target device 100 includes an abnormal sound. . This allows the user (including the administrator) to notice the abnormal state of the target device.

Further, the processor 301 detects sound data of the minimum stationary sound that minimizes the spectral power of the sound data of one or more frame periods determined to be the stationary sound during the learning mode, and detects the sound data of the detected minimum stationary sound. The sound data is stored in the storage 303 as normal sound sound data. Accordingly, in the learning mode, the information processing apparatus 30 can store the sound for one frame period, which is picked up by the microphone with the least disturbance sound, as the normal sound emitted by the target device. Therefore, the information processing apparatus 30 can obtain the normal sound to be compared with the least disturbance sound when identifying whether or not the sound emitted from the target device includes an abnormal sound.

In addition, the processor 301 detects the sound data of the minimum stationary sound that minimizes the spectral power of the sound data of one or more frame periods determined to be the stationary sound during the operation mode, and detects the sound data of the detected minimum stationary sound. Whether or not the sound emitted from the target device 100 includes an abnormal sound is identified according to the comparison between the sound data and the sound data of the normal sound. As a result, the information processing device 30 compares the current minimum stationary sound, in which disturbance sounds and the like are suppressed, with the minimum stationary sound stored as a normal sound in the operation mode, thereby determining whether the target device is in a normal state/abnormal state. The state identification accuracy can be improved.

(Embodiment 2)
In the first embodiment, normal/abnormal identification processing is performed using the minimum stationary sound among a plurality of stationary sounds extracted within a certain period of time. A case will be shown in which normal/abnormal identification processing is performed using the above steady sounds.

Further, in the abnormal noise determination system of the second embodiment, the same components as those of the first embodiment are denoted by the same reference numerals, and descriptions thereof are omitted.

FIG. 10 is a diagram showing an example of a schematic configuration of an abnormal noise determination system 5A according to the second embodiment. The abnormal sound determination system 5A includes two microphones 10A and 10B, an AD converter 20, an information processing device 30, and a cloud server 50. FIG. The two microphones 10A, 10B and the AD converter 20 are placed in a machine room in which the devices to be tested 100A, 100B are installed. The information processing device 30 is placed in a monitoring room where a user (including an administrator) is located.

FIG. 11 is a diagram showing the hardware configuration of the abnormal noise determination system 5A. The two microphones 10A and 10B have the same specifications as the microphone 10 in the first embodiment. The microphone 10A mainly picks up sounds emitted by the target device 100A. The microphone 10B mainly picks up sounds emitted by the target device 100B. Note that the number of microphones may be any number corresponding to the number of devices to be inspected.

The AD converter 20 analog-to-digital converts the sounds picked up by the two microphones 10A and 10B, and outputs digital sound data. This analog-to-digital conversion may be performed time-divisionally or simultaneously on the sound signals from the two microphones 10A and 10B.

The information processing device 30 is a terminal device connectable to the cloud server 50 and has the same configuration as that of the first embodiment. The information processing device 30 receives sound data output from the AD converter 20 via the network device 41 . The information processing device 30 accumulates sound data from the two microphones 10A and 10B in the storage 303 for a certain period of time (for example, 30 minutes, 1 hour, etc.). The information processing device 30 transmits the sound data for a certain period of time accumulated in the storage 303 to the cloud server 50 via the network device 45 . Note that the information processing apparatus 30 may transmit only sound data of a length required for normality/abnormality identification processing among the sound data for a certain period of time accumulated in the storage 303 . The information processing device 30 also functions as a display capable of displaying the results of normality/abnormality identification processing.

The cloud server 50 has a processor 501 , memory 502 , storage 503 and communication section 504 . The processor 501 has various processing devices such as a CPU (Central Processing Unit), a DSP (Digital Signal Processor), and an FPGA (Field Programmable Gate Array), and executes processing related to sound data. A memory 502 has a memory device such as a RAM (Random Access Memory), is used as a working memory of the processor 501, and is used for temporary storage in calculations during data processing. The memory 502 has a memory device such as a ROM (Read Only Memory), and stores various execution programs for executing the processing of the processor 301 and various setting data related to processing such as machine learning. The storage 503 has various storage devices such as HDD (Hard Disk Drive), SSD (Solid State Drive), and optical disk drive, and stores data such as sound data of the target device and learned models generated by machine learning. . The storage 503 is a large-capacity storage medium compared to the storage 303 incorporated in the information processing apparatus 30 in order to store sound data picked up in various environments. The communication unit 504 is an interface that performs wired or wireless communication, communicates with the information processing device 30 connected to the cloud NW via the network device 45, and transmits and receives data such as sound data and trained models.

The network device 41 relays data transmitted and received between the two microphones 10A and 10B connected to an intranet such as an in-house LAN (Local Area Network) and the information processing device 30 . Note that if the AD converter 20 has a network I/F (Interface), the network device 41 can be omitted.

The network device 45 relays data transmitted and received between the information processing device 30 and the cloud server 50 according to a communication protocol such as Ethernet (registered trademark).

Next, the operation of the abnormal noise determination system 5A will be described.

(learning mode)
The learning may be performed only once in the initial state in which the inspection target devices 100A and 100B are installed, or may be performed regularly or irregularly during maintenance or the like. In the sound data learning mode, the microphones 10A and 10B mainly pick up sounds emitted from the target devices 100A and 100B, respectively. Recording by the microphones 10A and 10B is performed for a predetermined time. The predetermined time is any fixed time such as 30 minutes, 1 hour, or 1 day. Also, the predetermined time may not be a fixed time but may be a time manually stopped by a user or an administrator.

FIG. 12 is a flow chart showing a learning operation procedure for sound data. The processing of steps S51 to S54 is the same as the processing of steps S1 to S4 in FIG. 3 of the first embodiment.

The processor 501 of the cloud server 50 receives sound data picked up by the microphones 10A and 10B and analog-to-digital converted by the AD converter 20 via the information processing device 30 and the network device 45 (S51).

Processor 501 accumulates the received sound data in storage 503 . The processor 501 determines whether or not a predetermined period of time has elapsed and recording by the microphone 10 has ended (S52). If recording has not ended, the processor 501 returns to the process of step S51.

When recording ends in step S52, the processor 501 performs frequency analysis on the time-series sound data accumulated in the storage 503 (S53). This frequency analysis is the same as in the first embodiment.

Based on the result of frequency analysis, the processor 501 performs stationary sound extraction processing for extracting stationary sounds contained in the sound data on a frame-by-frame basis (S54). This stationary sound extraction process is the same as that of the first embodiment.

The processor 501 accumulates all the sound data of one or more extracted stationary sounds in the storage 503 (S55). After this, the processor 501 ends this learning operation.

(operation mode)
In the operational mode, the processor 501 detects stationary sounds by the same processing as in the learning mode. The processor 501 compares the current stationary sound with one or more stationary sounds stored in the storage 503 when the devices 100A and 100B to be inspected are in a normal state. Processor 501 identifies the normal state/abnormal state of target devices 100A and 100B based on this comparison result.

FIG. 13 is a flow chart showing a procedure for operating sound data. Operational operations are performed regularly or irregularly. In the operation mode, the microphones 10A and 10B mainly pick up sounds emitted from the target devices 100A and 100B to be inspected.

In the operational operation, the processes of steps S61 to S64, S65 and S66 are the same as the processes of steps S11 to S14, S16 and S17 in FIG. 7 of the first embodiment.

The processor 501 of the cloud server 50 receives sound data picked up by the microphones 10A and 10B and analog-to-digital converted by the AD converter 20 via the information processing device 30 and the network device 45 (S61). Processor 501 accumulates the received sound data in storage 503 .

The processor 501 determines whether or not a certain period of time (for example, 30 minutes, 1 day, etc.) has passed since the microphones 10A and 10B started picking up sounds (S62). When the predetermined time has passed in step S62, the processor 501 performs frequency analysis on the time-series sound data accumulated in the storage 503 (S63). This frequency analysis is the same as in the learning mode.

Based on the result of frequency analysis, the processor 501 performs stationary sound extraction processing for extracting stationary sounds contained in the sound data on a frame-by-frame basis (S64). This stationary sound extraction processing is the same as in the learning mode.

The processor 501 performs normal/abnormal identification processing for identifying whether the target devices 100A and 100B to be inspected are in a normal state or an abnormal state based on one or more stationary sounds extracted in step S64 (S65). The normal/abnormal identification process is performed for each of the target devices 100A and 100B. The details of this normal/abnormal identification process will be described later.

The processor 501 determines whether or not to end the operational operation (S66). The end of the operational operation is recognized when the user (including the administrator) manually instructs to stop and the processor 301 accepts this instruction via the operation unit 305 . Processor 501 receives a stop instruction from information processing device 30 . Note that the operation operation may be terminated automatically by the processor 501 (for example, timer termination). In this case, manual operation by the user can be omitted. When not ending the operation operation, the processor 501 returns to the process of step S61 and repeats the same process. On the other hand, when ending the operational operation, the processor 501 ends this operational operation.

FIG. 14 is a flow chart showing the normal/abnormal identification procedure in step S65. The processor 501 reads sound data of one or more stationary sounds stored in the storage 503 in the learning mode (S71). The processor 501 calculates the degree of similarity between the current stationary sound data extracted in step S64 and the stationary sound data stored in the storage 503 (S72). This degree of similarity calculation is performed using a learned model and a cross-correlation function by machine learning.

The processor 501 determines whether or not there is at least one stationary sound whose similarity calculated in step S72 is equal to or greater than the threshold TH2 (S73). The threshold TH2 is a value for the processor 501 to determine whether or not the sound data of stationary sounds are similar. The threshold TH2 may be the same value as the threshold TH1, or may be a different value. If the similarity is less than the threshold TH2, the processor 501 determines that the inspection target devices 100A and 100B are emitting abnormal sounds (S74). The processor 501 performs notification processing (S75).

In this notification process, processor 501 transmits data indicating that at least one of target devices 100A and 100B is in an abnormal state to information processing device 30 via communication unit 504, cloud NW, and network device 45. FIG. When information processing apparatus 30 receives data indicating that at least one of target devices 100A and 100B is abnormal from cloud server 50, information processing apparatus 30 displays on display 306 that at least one of target devices 100A and 100B is in an abnormal state. . Note that if the cloud server 50 has a display, it may display that the target devices 100A and 100B are in an abnormal state on the display of its own device. After that, the processor 501 terminates this operational operation and returns to the main processing.

When the degree of similarity is equal to or greater than the threshold TH2 in step S73, the processor 501 determines that the target devices 100A and 100B to be inspected are emitting normal sounds (S76). In this case, since the target devices 100A and 100B to be inspected are in a normal state, the processor 501 does not perform notification processing. When target devices 100A and 100B to be inspected are in a normal state, processor 501 processes data indicating that target devices 100A and 100B are in a normal state via communication unit 504, cloud NW, and network device 45. It may be sent to device 30 . When information processing apparatus 30 receives data indicating that target devices 100A and 100B are in a normal state from cloud server 50, information processing apparatus 30 displays on display 306 that target devices 100A and 100B are in a normal state.

Thus, in the abnormal sound determination system 5A according to Embodiment 2, the processor 501 of the cloud server 50 accumulates one or more stationary sounds in the storage 503 during the learning mode. Processor 501, in the operation mode, if the stationary sound of the sound data picked up by microphones 10A and 10B is similar to any of a plurality of stationary sounds accumulated in storage 503, that is, if the stationary sound is similar to one or more of the stationary sounds If there is a stationary sound with a high degree of similarity, it is determined that the target devices 100A and 100B are in a normal state. Further, the cloud server 50 determines that the target devices 100A and 100B are in an abnormal state when there is no stationary sound with a high degree of similarity among the plurality of stationary sounds. Therefore, the cloud server 50 can identify the normality/abnormality of the target devices 100A and 100B when the target devices are placed in another environment and similar to any stationary sound.

As described above, the processor 501 stores two or more stationary sounds in the storage 503 as normal sounds from among the one or more one-frame period sound data determined to be stationary sounds during the learning mode. Thereby, the processor 501 of the cloud server 50 can store one or more stationary sounds in the storage 503 as normal sounds in the learning mode. Therefore, the processor 501 can obtain many normal sounds to be compared when identifying whether or not the sounds emitted from the target devices 100A and 100B include abnormal sounds. In this case, the cloud server 50 can omit the process of selecting the minimum stationary sound having the lowest spectral power from among the plurality of stationary sounds in the learning mode.

In addition, the processor 501 compares the sound data of one frame period determined to be a stationary sound during the operation mode with the sound data of two or more normal sounds stored in the storage 503, and determines whether the target devices 100A and 100B Identify whether or not the sound emitted from the device contains an abnormal sound. Thereby, the processor 501 of the cloud server 50 identifies that the target devices 100A and 100B are in a normal state when there is a stationary sound with a high degree of similarity among the plurality of stationary sounds. On the other hand, the processor 501 identifies that the target devices 100A and 100B are in an abnormal state when there is no stationary sound with a high degree of similarity among the plurality of stationary sounds. In this case, the processor 501 of the cloud server 50 can omit the process of selecting the minimum stationary sound having the lowest spectral power from among the plurality of stationary sounds in the operation mode.

Note that the cloud server learns the minimum stationary sound from among the plurality of extracted stationary sounds as a normal sound, and compares the current minimum stationary sound with the normal sound in the operation mode, as in the first embodiment. Thus, the normality/abnormality of the target devices 100A and 100B may be identified.

Moreover, although the case where the cloud server 50 performs the learning operation and the operating operation has been described in the second embodiment, the information processing device 30 may perform the same learning operation and operating operation as the cloud server 50 does. In this case, the cloud server 50 can be omitted.

(Modification of Embodiment 2)
In addition, during the learning mode, the processor 501 of the cloud server 50 accumulates normal sounds picked up by the microphones 10 and 10B in various environments (for example, various places where the target device is installed) in the storage 503. good. During the operation mode, the processor 501 of the cloud server 50 can identify the normal/abnormal states of the target devices 100A and 100B in various environments using a large amount of normal sound data accumulated in the storage 503 .

In this way, the storage 503 stores sound data for a predetermined period of sounds emitted from the target devices 100A and 100B placed in a plurality of different environments, and determines that the sound is stationary during the learning mode for each environment. At least one stationary sound is stored as a normal sound from among the sound data of one or more one frame period. The processor 501 compares the sound data of one frame period determined to be a steady sound during the operation mode different from the learning mode with the sound data of the normal sound stored for each environment in the storage 503, and determines whether the target device 100A , 100B includes an abnormal sound.

As a result, even when the target devices are placed in various environments, the cloud server 50 can identify the abnormal state or normal state of the target devices 100A and 100B corresponding to various environments.

Although the embodiments have been described above with reference to the drawings, it goes without saying that the present disclosure is not limited to such examples. It is obvious that a person skilled in the art can conceive of various modifications, modifications, substitutions, additions, deletions, and equivalents within the scope of the claims. Naturally, it is understood that it belongs to the technical scope of the present disclosure. Also, the components in the above-described embodiments may be combined arbitrarily without departing from the spirit of the invention.

The present disclosure provides an anomaly determination device for determining whether or not an abnormal sound is included in a sound picked up from a target device, even when the sound picked up from a target device includes not only a stationary sound but also an external noise such as a sudden sound. It is useful because it can improve identification accuracy.

5 Abnormal Sound Determination System 10, 10A, 10B Microphone 20 AD Converter 30 Information Processing Device 50 Cloud Server 100 Target Equipment 301, 501 Processor 302, 502 Memory 303, 503 Storage 304, 504 Communication Unit 305 Operation Unit 306 Display 307 Input Unit

Claims (9)

an input unit connected to a microphone that picks up sound emitted from the target device;
a memory for storing sound data for a predetermined period of sound emitted from the target device;
a processor that uses the sound data for the predetermined period and determines whether the sound corresponding to the sound data for the one frame period is a stationary sound or a sudden sound for each one frame period shorter than the predetermined period,
The memory stores at least one stationary sound as a normal sound from among the one or more of the one-frame-period sound data determined to be the stationary sound during the learning mode;
The processor compares the sound data of the one-frame period determined to be the stationary sound during an operation mode different from the learning mode with the sound data of the normal sound stored in the memory, and performs the target identify whether or not the sound emitted from the device contains abnormal noise;
Abnormal sound judgment device. When the processor identifies that the sound emitted from the target device includes an abnormal sound, the processor displays on the display an abnormal sound detection notification indicating that the sound emitted from the target device includes the abnormal sound.
The abnormal noise determination device according to claim 1. The processor detects sound data having a minimum power of one or more of the sound data of the one-frame period determined as the stationary sound during the learning mode, and converts the detected sound data to the sound of the normal sound. store as data in said memory;
The abnormal noise determination device according to claim 1. The processor detects sound data having a minimum power among the one or more of the one-frame-period sound data determined to be the stationary sound during the operation mode, and detects the detected sound data and the normal sound. Identifying whether or not the sound emitted from the target device includes an abnormal sound according to the comparison with the data;
The abnormal noise determination device according to claim 3. The processor stores, in the memory, two or more stationary sounds from among the one or more of the one-frame-period sound data determined as the stationary sounds during the learning mode, as the normal sounds.
The abnormal noise determination device according to claim 1. The processor compares the one-frame-period sound data determined to be the stationary sound during the operation mode with the sound data of the two or more normal sounds stored in the memory. identify whether or not the sound emitted from the
The abnormal noise determination device according to claim 5. The memory stores sound data for a predetermined period of sounds emitted from the target device placed in a plurality of different environments, and at least one sound determined to be the stationary sound during the learning mode for each of the environments. storing at least one stationary sound as a normal sound from the sound data of the one frame period of
The processor compares the sound data of the one-frame period determined as the stationary sound during an operation mode different from the learning mode with the sound data of the normal sound stored in the memory for each environment. to identify whether or not the sound emitted from the target device includes an abnormal sound;
The abnormal noise determination device according to claim 1. An abnormal noise determination method executed by an abnormality determination device,
a step of inputting sound data for a predetermined period of sound emitted from the target device from a microphone that picks up the sound emitted from the target device;
a step of determining whether the sound corresponding to the sound data for the one-frame period is a stationary sound or a sudden sound for each one-frame period shorter than the predetermined period, using the input sound data for the predetermined period;
acquiring at least one stationary sound stored as a normal sound from among the one or more frame period sound data determined as the stationary sound during the learning mode;
According to a comparison between the sound data of the one-frame period determined to be the stationary sound and the sound data of the normal sound acquired during the operation mode different from the learning mode, the sound emitted from the target device is different. and identifying whether sounds are included.
Abnormal sound judgment method. an input unit connected to a microphone that picks up sound emitted from the target device;
a memory for storing sound data for a predetermined period of sound emitted from the target device;
A controller that uses the sound data for the predetermined period and determines whether the sound corresponding to the sound data for the one frame period is a steady sound or a sudden sound for each one frame period shorter than the predetermined period,
The memory stores at least one stationary sound as a normal sound from among the one or more of the one-frame-period sound data determined to be the stationary sound during the learning mode;
The controller compares the sound data of the one-frame period determined to be the stationary sound during an operation mode different from the learning mode with the sound data of the normal sound stored in the memory, and performs the target identify whether or not the sound emitted from the device contains abnormal noise;
Abnormal sound judgment system.

Images