Abnormal sound source localization method and device based on sound pressure amplitude ratio

Abstract

The present invention discloses a method and device for locating abnormal sound sources based on the sound pressure amplitude ratio. A rectangular space is planned with the target area to be detected as the center, and a group of audio acquisition devices are arranged at each vertex of the rectangular space. The positioning method includes normal state audio data collection and training, abnormal audio recognition and extraction, primary sound source location positioning, and secondary sound source location positioning. After the sound waves pass through the sensor array in sequence, the sequence and characteristics are analyzed, and the direction and distance of the noise are calculated. Then, in an environment with a lot of noise, the location of the fault can be accurately found. At the same time, the non-contact nature of the sound signal can effectively avoid the difficulty of collecting vibration signal data. The sound source is located by utilizing the difference in the intensity of the sound signals from the same sound source received by different microphones.

Description

Technical Field

[0001] The invention relates to a method and a device for detecting a fault position by using audio. Background Art

[0002] Sound source localization plays a crucial role in sound signal processing and is widely used in applications such as smart devices, video conferencing systems, traffic violation capture, and fault diagnosis, enabling automatic capture and alignment of sound sources. This method processes the collected signal to determine the direction of arrival of the sound source at the entire microphone array. Compared to a single microphone sensor, a microphone array composed of multiple microphones offers superior advantages in speech signal processing. Microphone arrays are complementary and can effectively eliminate background noise. This paper exploits the differences in the intensity of sound signals from the same source received by different microphones to achieve sound source localization.

[0003] In current industrial manufacturing, we have found that as the scale of the power grid expands, the workload of equipment maintenance and testing has increased dramatically. As the economic and social costs brought about by equipment maintenance have become increasingly prominent, traditional equipment testing and detection methods have restricted the development of the power grid and companies. Summary of the Invention

[0004] The technical problem to be solved by the present invention is: to address the problem that it is difficult to quickly find the source of a fault during equipment maintenance and testing work, and to provide a sound source localization method and device based on sound pressure amplitude ratio to determine the specific position of the sound source in space.

[0005] The technical solution of the present invention is:

[0006] A method for locating an abnormal sound source based on a sound pressure amplitude ratio, characterized by comprising the following steps:

[0007] S1. Normal state audio data collection and training: Arrange multiple audio collection devices to collect audio data of the target area, first collect and train the audio model under normal working conditions;

[0008] S2. Abnormal Audio Identification and Extraction: The audio acquisition device collects audio data from the target area in real time, compares it with the trained audio model, and identifies abnormal frequencies;

[0009] S3. A sound source location: Compare the abnormal frequency intensity collected by each audio collection device, select the audio collection device with the largest intensity as the main device, and the abnormal sound source is closest to the main device;

[0010] S4. Secondary location of the sound source: Perform an inverse Fourier transform on the abnormal frequency spectrum data, extract the amplitude of the abnormal sound source audio data in the time domain obtained by the inverse Fourier transform as the sound pressure amplitude of the abnormal audio, combine the abnormal frequency sound pressure amplitudes obtained by each audio acquisition device, and use the relationship between the pressure-amplitude ratio and the inverse proportion of the square of the distance to perform secondary location of the abnormal sound source and determine the actual coordinates of the abnormal sound source.

[0011] In S1: the target area is a rectangular space, and the generator set is located in the space. A group of audio collection devices is arranged at each of the six vertices of the rectangular space, and each group of audio collection devices consists of four microphones at different positions.

[0012] In S1: The collected audio data is transformed through Fourier transform to obtain a spectrum diagram, and the main frequency of the audio in the frequency domain is found through peak detection. The above operation is performed on multiple groups of audio data to obtain the main frequency, which is used as the feature input into the isolation forest machine learning algorithm for training to obtain the corresponding model.

[0013] In S2: Abnormal signal identification and extraction are performed on the collected audio data. The audio signal to be identified is Fourier transformed to obtain its frequency domain information. Peak detection is then performed to obtain the main frequency. Finally, the main frequency is input as a feature into the trained machine learning model. After the model determines the input feature and obtains the abnormal frequency, the audio signal is bandpass filtered to obtain the time domain information of the abnormal frequency.

[0014] In S3: the real-time collected audio data is compared with the trained normal state model data. If the normal state model identifies a new frequency in the frequency domain of the audio data, and the intensity of the new frequency is greater than the specified threshold, the new frequency is regarded as an abnormal frequency, and the abnormal frequency is extracted by filtering in the frequency domain.

[0015] In S4: The sound pressure amplitude ratio equation is constructed using the other three microphones of the main microphone and the four nearest adjacent microphones, and the optimal estimated position of the sound source is found by minimizing the residual error of the equation.

[0016]

[0017] Among them, H is the minimum residual sum of the equation, r is the distance from the sound source to the microphone, E is the effective sound pressure, It is the sound pressure amplitude ratio, that is, the effective sound pressure ratio, and its value is obtained from the actual measurement values ​​of each microphone.

[0018] After S4, S5 is executed: the position coordinates of the abnormal audio signal are displayed on a two-dimensional image as the position of the abnormal sound source.

[0019] An abnormal sound source localization device based on sound pressure amplitude ratio includes an audio collection device. A rectangular space is planned with the target area to be detected as the center, and a group of audio collection devices are arranged at each vertex of the rectangular space. All audio collection devices are connected to a control module through a switch.

[0020] Build a bracket to install the audio collection device. Each set of audio collection devices consists of 2-6 microphones. Each set of microphones is arranged in a straight line. The four sets of microphones located at the lower vertex of the cuboid are arranged in a straight line along the direction of gravity, and the four sets of microphones located at the upper vertex are arranged in a straight line along the horizontal direction.

[0021] The control module is an industrial control computer used to control the automatic initialization, data acquisition, data transmission and calculation of the entire system. At the same time, the industrial computer is equipped with an alarm system and a display module to broadcast the operating status of the entire system.

[0022] The beneficial effects of the present invention are:

[0023] Most mechanical equipment emits a steady, regular noise during normal operation. However, when equipment ages or experiences other malfunctions, it produces significantly different operating noise than normal. This allows for troubleshooting by locating the source of the abnormal noise. Abnormal equipment noise often manifests as an audible sound (20-20 kHz). Arcing or gas leaks generate ultrasonic waves with frequencies higher than audible. Therefore, by measuring the ultrasonic component in noisy environments, gas leaks and arcing can be detected. Sound waves are directional, making them easy to detect. After the sound waves pass through an array of sensors, their sequence and characteristics are analyzed to calculate the direction and distance of the noise. This allows pinpointing the location of the fault even in noisy environments. Furthermore, the non-contact nature of sound signals effectively mitigates the difficulties of collecting vibration signal data. Sound source localization is achieved by utilizing the intensity differences in sound signals received by different microphones from the same source. The proposed algorithm directly measures the sound pressure amplitude ratio. This system allows for flexible adjustments to position coordinate errors based on practical considerations, such as computation time and the amount of data to be processed. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 It is a logic flow chart of the present invention.

[0025] Figure 2 It is a schematic diagram of the device of the present invention. DETAILED DESCRIPTION

[0026] Example 1:

[0027] The abnormal sound source location method of the generator set based on the sound pressure amplitude ratio includes:

[0028] S1. Audio data acquisition and training for normal unit conditions: The audio acquisition module consists of a three-dimensional microphone matrix consisting of eight acquisition cards and eight four-element linear microphone arrays. The 32-channel microphones collect audio data during normal generator set operation, and deep learning is used to train the audio model for normal unit conditions.

[0029] S2. Abnormal audio identification and extraction: During actual monitoring, 32 microphones simultaneously collect audio data from the unit's real-time operation. The real-time audio data collected by the 32 microphones is received by the industrial computer and compared with the trained normal state model data of the unit. If a new frequency is identified in the frequency domain of the audio data by comparison with the normal state model, and the intensity of the new frequency is greater than a specified threshold (the threshold is determined by the median value of the audio data in the frequency domain under normal conditions), the new frequency is considered an abnormal frequency and is extracted by filtering in the frequency domain.

[0030] S3. Determine the location of the sound source: First, compare the intensity of the abnormal frequency of the abnormal sound source extracted by the 32 microphones in the frequency domain. The microphone with the largest intensity is used as the main microphone, and the abnormal sound source is closest to the microphone position, so as to perform "rough" positioning of the abnormal sound source; then, perform an inverse Fourier transform on the abnormal frequency spectrum data after filtering in the previous step, extract the amplitude of the abnormal sound source audio data obtained by the inverse Fourier transform in the time domain as the sound pressure amplitude of the abnormal audio, and combine the abnormal frequency sound pressure amplitudes obtained by the 32 microphones in pairs. Use the relationship that the pressure-amplitude ratio is inversely proportional to the square of the distance to perform a second "fine" positioning of the abnormal sound source, that is, determine the actual coordinates (x, y, z) of the abnormal sound source;

[0031] S4. Visualize the detected abnormal sound source location on a two-dimensional unit image.

[0032] Example 2:

[0033] The method for locating abnormal sound sources of generator sets based on sound pressure amplitude ratio includes audio data collection, recognition, extraction and processing, abnormal audio and video signal alarm, abnormal sound source signal estimation and positioning, and abnormal sound source signal visualization. The specific steps are as follows:

[0034] S1: 8 groups of 4-element microphones, or a 32-channel microphone array, are used to collect audio data from the generator set in real time. The collected audio files are Fourier transformed to obtain a spectrogram. Peak detection is used to find the main frequency of the audio in the frequency domain. The above operation is performed on multiple groups of audio files to obtain the main frequency. This is used as the feature input into the isolation forest machine learning algorithm for training to obtain the corresponding model. After the audio model of the generator set under normal conditions is trained, the audio data collected on site is compared with the trained audio model to determine whether the generator set generates abnormal audio signals.

[0035] S2: Feature engineering is performed by collecting a large amount of normal generator set audio data. This involves training a model based on the dominant frequencies in the spectrogram. Abnormal frequencies are detected using the isolation forest algorithm. After performing an inverse Fourier transform and filtering, their time domain information is extracted. The collected audio data is then used for abnormal signal identification and extraction. The audio signal to be identified undergoes a Fourier transform to obtain its frequency domain information. Peak detection is then performed to determine the dominant frequencies, which are then input as features into the trained machine learning model. The model evaluates the input features, outputting a 1 if normal and a -1 if abnormal. Once the abnormal frequencies are determined, the audio signal is bandpass filtered to obtain its time domain information.

[0036] S3: Compare the abnormal audio signals collected by the 32 microphones, find the microphone with the largest amplitude and determine it as the main microphone to achieve "coarse" positioning. Based on this, perform secondary "fine" positioning, that is, determine the coordinate position (x, y, z) of the abnormal audio signal;

[0037] For simplicity, let's first consider the problem of sound source localization on a two-dimensional plane. Assume that four microphones are equally spaced on the x-axis, with coordinates (-3a, 0), (-a, 0), (a, 0), and (3a, 0). If the sound source is located at point S(x, y), the distances from the sound source to the four microphones can be expressed as:

[0038]

[0039]

[0040]

[0041]

[0042] According to the relevant knowledge about sound waves, the sound pressure generated by the sound source at the i microphones is given by the following formula,

[0043]

[0044] Where p0 is the voltage generated by the sound source at the reference distance, r0, and ω is the angular frequency of the simple harmonic vibration of the sound source. c0 is the speed of sound, and Δ=r i -r0.

[0045] According to the characteristics of the microphone, the voltage output generated by the above sound pressure at the i-th microphone is:

[0046] e i (t) = p i (t)H i cosθ i

[0047] Among them, H i is a parameter determined by the transfer characteristic of the i-th microphone, θ i is the incident angle of the sound wave at the i-th microphone. If the transfer characteristics of the four microphones are exactly the same (represented by H), then for any two of them, the received signals e i (t) and e j (t), except for the different amplitudes and a fixed time delay Δt ij =k(r i -r j ), the waveforms are the same.

[0048] Next, we derive several parameters related to sound source localization. For this purpose, we have the ratio of e1(t) to e2(t):

[0049]

[0050] From the above, we can see that

[0051] e2(t+Δt 12 )=p2(t+Δt 12 )Hcosθ2

[0052] =p2(t)Hcosθ2exp{jk(t2-t1)}

[0053] =p2(t)Hcosθ2exp{jk(r2-r1)}

[0054] so,

[0055]

[0056] Similarly, there is a similar relationship between e3(t) and e4(t),

[0057]

[0058] Assume that the effective sound pressure of a sound source is E1, E2, E3, and E4, which are the effective sound pressures received by the sound pressure sensor, and the distances from the sound source to the sensor are r1, r2, r3, and r4, respectively. The following formula can be obtained:

[0059]

[0060] in and is the sound pressure amplitude ratio, i.e., the effective sound pressure ratio, which can be obtained from the actual measurement values ​​of each microphone;

[0061] The "coarse" positioning process includes: comparing the sound pressure amplitudes of the abnormal audio data extracted by 32 microphones, and taking the position of the microphone with the largest amplitude as the direction of the abnormal sound source to "coarsely" locate the abnormal sound source, that is, determine the main microphone;

[0062] The "fine" positioning process includes: constructing a sound pressure amplitude ratio equation using the other three microphones of the main microphone and the four nearest neighboring microphones, finding the optimal estimated location of the sound source by minimizing the residual sum of the equation, and substituting the coordinate set of the "coarse" positioning area into the estimated coordinates (x, y, z) of the abnormal audio position, thus performing "fine" positioning of the abnormal sound source. Based on the research results of this article, the preliminary abnormal audio location estimate is as follows:

[0063]

[0064] Among them, H is the minimum sum of the residuals of the equation;

[0065] For example, in one test, a sound source emitting an abnormal frequency was placed at the spatial coordinates (210, 170, 120). The sound pressure amplitude of the collected abnormal audio data was used for "rough" positioning, and the main microphone number was found to be 3. According to the positional relationship of the microphones, the nearest adjacent microphone is microphone 4. At this time, the amplitude extracted from the first microphone of microphone 3 is recorded as E1, and the coordinate matrix of the spatial area determined by the rough positioning is recorded as D. The coordinates of the main microphone are (265, 80.7, 179), and the coordinates of the remaining seven microphones are (265, 84.9, 179), (265, 89.1, 179), (265, 93.3, 179), (265, 93.3, 21), (265, 89.1, 21), (265, 84.9, 21), and (265, 80.7, 21). Expand the coordinates of the eight microphones into a matrix of the same size as the coordinate matrix D and substitute it into the formula

[0066]

[0067] Where A represents the expanded microphone coordinate matrix, and the distance matrix r1, r2, r3, r4, r5, r6, r7, r8 from the spatial coordinates to the 8 microphones is calculated. Then the distance matrix and the amplitude extracted from each microphone are substituted into the formula

[0068]

[0069] The minimum residual sum H is calculated to be 227114.51. Finally, the coordinate position of H is found by indexing, which is the calculated abnormal audio position (213, 171, 121).

[0070] S4: Display the abnormal sound source position on a two-dimensional image using the position coordinates (x, y, z) of the abnormal audio signal.

[0071] Embodiment 3: The sound source localization device includes:

[0072] The first module, the equipment module, includes industrial computers (IPCs), specifically designed for industrial sites. They are a general term for tools used to monitor and control production processes, electromechanical equipment, and process equipment. These computers possess important computer attributes and characteristics. They utilize a bus architecture and are primarily used for monitoring and controlling production processes, electromechanical equipment, and process equipment.

[0073] A switch's main function is to connect network devices and transmit data through data exchange. A network switch is a device that expands a network and provides more connection ports for a sub-network to connect more network devices.

[0074] A display screen is a display used in industrial control processes or equipment. It is quite different from civilian or commercial displays and has special designs such as dustproof and shockproof.

[0075] Alarm: In order to prevent danger from causing unnecessary consequences and losses, an alarm is installed to monitor the generator set in real time to see if there is any abnormality.

[0076] Metal brackets were built around the entire generator set to hold the eight microphones. The generator set measures 4000mm in length, 3000mm in width, and 250mm in height. Therefore, two 4000mm-long metal brackets were placed on either side of the generator set, one 3000mm-long bracket was placed at the rear, and four 2500mm-long brackets were placed vertically along the generator set. Seven brackets in total.

[0077] The second module, the acquisition module, is used to acquire the input sound signals from each microphone in the microphone array. As described in this article, eight acquisition cards and eight four-element linear microphone arrays were selected. The 32-channel microphone array was arranged into a three-dimensional structure based on the actual size of the generator set. The four bottom microphones in the array were 500mm above the ground to avoid interference. Each microphone group was powered by a 12V two-hole power strip. In the three-dimensional model, the array can receive incident signals from all directions in the space.

[0078] The third module, the monitoring module, transmits the audio data collected by the microphone array to the industrial computer via a switch. The industrial computer centrally controls the automatic initialization, data collection, and data transmission of the entire system. The industrial computer can also be equipped with an alarm system to broadcast the operating status of the entire system in real time. UDP / IP broadcasts the raw stream signal in real time, and TCP / IP receives the signal.

[0079] The fourth module, the display module, uses the display screen to display two-dimensional abnormal audio signals. The alarm light displays different colors to represent the system operation status. If an abnormal state is identified, an audible and visual alarm will be issued through the alarm. When the generator set is operating normally, the alarm will display green. If abnormal audio is detected, the yellow light will flash and the buzzer will sound an alarm. If abnormal video such as smoke is detected, the yellow light will flash and the buzzer will sound an alarm.

[0080] This invention is used on generator sets or equipment with relatively high noise levels. First, audio data from the normal operation of the equipment is collected to train an audio model of the normal state of the unit. During actual testing, the audio collected on site is compared with the trained audio model under normal conditions to determine whether the unit has entered an abnormal state. If an abnormal state is identified, an audible and visual alarm is issued through an alarm. If an abnormal state is identified, the frequency extraction and location of the abnormal sound source are continued. First, new frequencies are identified in the frequency domain of the audio data, and the new frequencies are regarded as abnormal frequencies. The abnormal frequencies are extracted through filtering, and the amplitude of the filtered data in the time domain is extracted through inverse Fourier transform as the sound pressure amplitude of the abnormal audio. By comparing the sound pressure amplitudes of the abnormal audio data captured by 32 microphones, the abnormal sound source is "roughly" located, using the position of the microphone with the largest amplitude as the direction of the abnormal sound source. A sound pressure amplitude ratio equation is constructed using the other three microphones and the four nearest neighboring microphones. The residual sum of the equation is minimized to find the optimal estimated sound source location. The coordinate set of the "roughly" located area is substituted into the estimated (x, y, z) to perform "fine" positioning of the exceeded sound source. Finally, the (x, y) coordinate of the given (x, y, z) is used to display the abnormal sound source location on a 2D image (this requires matching the 3D detection coordinate system with the 2D image display coordinate system).

Claims

1. A method for locating abnormal sound sources based on sound pressure amplitude ratio, characterized in that The following steps are involved: S1. Normal-State Audio Data Collection and Training: Multiple audio acquisition devices are deployed to collect audio data from the target area. First, an audio model is acquired and trained under normal working conditions. The target area is a rectangular space containing a generator set. A set of audio acquisition devices is deployed at each of the eight vertices of the rectangular space. Each set of audio acquisition devices consists of four microphones at different locations. S2. Abnormal Audio Identification and Extraction: The audio acquisition device collects audio data from the target area in real time, compares it with the trained audio model, and identifies abnormal frequencies; S3. One-time location of the sound source: Compare the abnormal frequency intensity collected by each audio collection device, select the audio collection device with the largest intensity as the main device, and the abnormal sound source is closest to the main device; S4. Secondary location of the sound source: perform inverse Fourier transform on the abnormal frequency spectrum data, extract the amplitude of the abnormal sound source audio data in the time domain obtained by the inverse Fourier transform as the sound pressure amplitude of the abnormal audio, combine the abnormal frequency sound pressure amplitudes obtained by each audio acquisition device, and use the relationship that the pressure-amplitude ratio is inversely proportional to the square of the distance to perform secondary location of the abnormal sound source and determine the actual coordinates of the abnormal sound source; construct a sound pressure amplitude ratio equation using the other 3 microphones of the main microphone and the 4 nearest adjacent microphones, and use the minimum sum of the residuals of the equation to find the optimal estimated position of the sound source Among them, H is the minimum residual sum of the equation, r is the distance from the sound source to the microphone, E is the effective sound pressure, It is the sound pressure amplitude ratio, that is, the effective sound pressure ratio, and its value is obtained from the actual measurement values ​​of each microphone.

2. The abnormal sound source localization method based on sound pressure amplitude ratio according to claim 1 is characterized in that In S1: The collected audio data is transformed through Fourier transform to obtain a spectrum diagram, and the main frequency of the audio in the frequency domain is found through peak detection. The above operation is performed on multiple groups of audio data to obtain the main frequency, which is used as the feature input into the isolation forest machine learning algorithm for training to obtain the corresponding model.

3. The abnormal sound source localization method based on sound pressure amplitude ratio according to claim 2 is characterized in that In S2: Abnormal signal identification and extraction are performed on the collected audio data. The audio signal to be identified is Fourier transformed to obtain its frequency domain information. Peak detection is then performed to obtain the main frequency. Finally, the main frequency is input as a feature into the trained machine learning model. After the model determines the input feature and obtains the abnormal frequency, the audio signal is bandpass filtered to obtain the time domain information of the abnormal frequency.

4. The abnormal sound source localization method based on sound pressure amplitude ratio according to claim 3 is characterized in that In S3: the real-time collected audio data is compared with the trained normal state model data. If the normal state model identifies a new frequency in the frequency domain of the audio data, and the intensity of the new frequency is greater than the specified threshold, the new frequency is regarded as an abnormal frequency, and the abnormal frequency is extracted by filtering in the frequency domain.

5. The abnormal sound source localization method based on sound pressure amplitude ratio according to any one of claims 1 to 4, characterized in that After S4, S5 is executed: the position coordinates of the abnormal audio signal are displayed on a two-dimensional image as the position of the abnormal sound source.

6. An abnormal sound source localization device based on sound pressure amplitude ratio, used to execute the abnormal sound source localization method based on sound pressure amplitude ratio according to claim 1, comprising an audio acquisition device, characterized in that: A rectangular space is planned with the target area to be detected as the center, and a group of audio collection devices are arranged at each vertex of the rectangular space. All audio collection devices are connected to the control module through a switch.

7. The abnormal sound source localization device based on sound pressure amplitude ratio according to claim 6, characterized in that: Build a bracket to install the audio collection device. Each set of audio collection devices consists of 2-6 microphones. Each set of microphones is arranged in a straight line. The four sets of microphones located at the lower vertex of the cuboid are arranged in a straight line along the direction of gravity, and the four sets of microphones located at the upper vertex are arranged in a straight line along the horizontal direction.

8. The abnormal sound source localization device based on sound pressure amplitude ratio according to claim 7, characterized in that: The control module is an industrial control computer used to control the automatic initialization, data acquisition, data transmission and calculation of the entire system. At the same time, the industrial computer is equipped with an alarm system and a display module to broadcast the operating status of the entire system.

Images